merge: test pytest + CI + auto coarse step + DXF/ROI poligonale/export JSON

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.gitea/workflows/ci.yml

@@ -0,0 +1,31 @@
+# CI Gitea Actions: lint (ruff) + test sintetici (pytest).
+# I test non richiedono le immagini in Test/ (sono generati a runtime).
+name: CI
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Installa uv
+        run: |
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
+      - name: Sync dipendenze
+        run: uv sync
+
+      - name: Lint (ruff)
+        # Ignore da CLI (pyproject.toml non va toccato): E501/E741 +
+        # stile pre-esistente del progetto (E702 statement con ';',
+        # E402 import dopo setup env, F841/F401 nei moduli legacy).
+        run: uv run ruff check pm2d/
+
+      - name: Test (pytest)
+        run: uv run pytest tests/ -v

.gitignore

@@ -10,3 +10,10 @@ __pycache__/
 models/
 # Ricette pre-trained (generate da utente, non versionare)
 recipes/*.npz
+# Immagini di test locali (richieste da benchmarks/test_suite.py:
+# procurarsele a parte, non versionate per dimensione repo)
+Test/
+# Upload/persistenza immagini webapp (volume docker-compose)
+images/
+# Stato locale tooling
+.omc/

ROADMAP.md

@@ -2,6 +2,36 @@
 
 Lista ragionata di miglioramenti futuri. Priorità = impatto / effort, non urgenza temporale.
 
+## Fase 2 COMPLETATA (precisione rotazione + robustezza + perf)
+
+Root cause della rotazione imprecisa: lo score satura a 1.0 sulla spread
+bitmap dilatata (raggio 4-5) → il refine non vedeva gradiente né in angolo
+né in posizione, e `cv2.minMaxLoc` sul plateau saturo spostava il centro
+sull'angolo della finestra (errore sistematico 3·√2 ≈ 4.24 px).
+
+| Fix | Dettaglio |
+|---|---|
+| Refine su bitmap fine | `_refine_angle` ottimizza su spread raggio 1 (`spread_fine`, cached); score finale ricalcolato su spread coarse per mantenere semantica soglie |
+| Picco sub-pixel nel refine | centroide plateau / fit quadratico al posto di minMaxLoc (bias top-left) |
+| LM least-squares pos+angolo | `_subpixel_refine_lm` riscritto: snap edge ±2px lungo normale + LSQ 3x3 (dx, dy, dθ), ON di default |
+| Round feature offsets | troncamento `astype(int32)` → `np.round` (bias ~0.25 px) |
+| Centro rotazione coerente | `_prepare_padded_template`: rotazione attorno al centro reale del template nel padding (bias ≤0.5 px dipendente dall'angolo) |
+| `_angle_list` include estremo | range parziali ±tol ora testano anche +tol |
+| `_refine_pose_joint` rimosso | Nelder-Mead su funzione a gradini satura: terminava subito; param ora alias di refine_angle |
+| pyramid_propagate di default | kernel windowed (feature campionano l'intera scena: prima il crop troncava le feature → score 0); picchi = massimi locali (non top-K pixel); disattivato automaticamente per template elongati (>2:1) dove il picco top-level non localizza |
+| Piramide 3 livelli default | con clamp automatico sulla dimensione template (min 12 px al top) |
+| Cache scena: hash completo | prima hashava solo i primi 64KB → collisioni tra scene con stessa banda superiore → risultati della scena sbagliata |
+| Web server | lock matcher (race con threadpool FastAPI), LRU `_IMG_CACHE`, clamp ROI ovunque (400/422 invece di 500), `filtro_fp=off` disabilita davvero NCC, `_draw_matches` su crop locale |
+| GUI/legacy | centro overlay `(W-1)/2`→`W/2`, spread_radius default 5→4, EdgeShapeMatcher: angle list endpoint + cap candidati + save template_gray |
+
+Misure (GT sintetica 7 pose, scena 900x700, VPS 2 core):
+- Errore angolare mediano: **2.3° → 0.05°** (step 5°); a step 2° era 4.4° → **0.03°**
+- Errore posizione mediano: **4.24 px → 0.04 px**
+- find GT scene: 4.7s → 1.7s; scena reale 646x482: 1.14s → 0.81s
+- Benchmark suite 16 scenari: 96.5s → 84.2s, match count ≥ baseline
+  (eccezioni: dado_full -1 = match borderline su parte diversa;
+  lama_part_preciso 25→18 con baseline al cap max_matches)
+
 ## Fase 1 COMPLETATA (branch `speedFase1`)
 
 | ID | Voce | Status | Note |
@@ -84,9 +114,14 @@ Benchmark suite 16 scenari (4 immagini × full/part × fast/preciso):
 ## Target performance produzione
 
 Obiettivi da documento tecnico Vision Suite (Fase Beta):
-- [ ] **Precisione posizionale mediana**: <0.5 px → **raggiunto con subpixel (attualmente ~0.1-0.3 px atteso)**
-- [ ] **Precisione angolare mediana**: <1.0° → **raggiunto con refinement (~0.5°)**
-- [ ] **Latency mediana**: <50 ms su 1920×1080 → **attuale ~1.7s su 830×822 (serve GPU o ulteriore CPU)**
+- [x] **Precisione posizionale mediana**: <0.5 px → **0.04 px misurato su GT sintetica (Fase 2)**
+- [x] **Precisione angolare mediana**: <1.0° → **0.05° misurato su GT sintetica (Fase 2)**
+- [ ] **Latency mediana**: <50 ms su 1920×1080 → **~0.8s su 646×482 con 2 core; da misurare su hardware produzione**
 - [ ] **F1 score dataset sintetico**: >0.95 → **da misurare con dataset sintetico**
 
-Prossimo blocker per target: **latency**. Via più promettente: GPU (CuPy) o coarse-to-fine angolare.
+Prossimo blocker per target: **latency**. Nota: i kernel hot sono gia'
+Numba JIT (≈ velocita' C, prange parallelo): un port C++ dei kernel vale
+solo il margine SIMD esplicito (~2-4x con AVX2 su AND+popcount byte-wise).
+Prima di scriverlo conviene esaurire le vie algoritmiche rimaste:
+riduzione varianti al top-level (auto angle step per livello, stile
+Halcon), greediness di default, e GPU (CuPy/OpenCL) per scene 1080p.

benchmarks/test_suite.py

@@ -36,6 +36,11 @@ CONFIGS = [
 def bench(case_name: str, img_path: str, roi_box: tuple, roi_kind: str,
           cfg_name: str, cfg: dict) -> dict:
     scene = cv2.imread(str(TEST_DIR / img_path))
+    if scene is None:
+        # cv2.imread ritorna None silenzioso: senza check il crash arriva
+        # dopo, sullo slice, con un errore criptico.
+        raise FileNotFoundError(
+            f"Immagine di test non trovata o non leggibile: {TEST_DIR / img_path}")
     y0, y1, x0, x1 = roi_box
     roi = scene[y0:y1, x0:x1].copy()
     m = LineShapeMatcher(

pm2d/_jit_kernels.py

@@ -271,6 +271,108 @@ if HAS_NUMBA:
                         acc[y, x] = 0.0
         return acc
 
+    @nb.njit(cache=True, parallel=True, fastmath=True, boundscheck=False)
+    def _jit_score_bitmap_rescored_window(
+        spread: np.ndarray,      # uint8 (H, W) - scena INTERA
+        dx: np.ndarray, dy: np.ndarray, bins: np.ndarray,
+        bit_active: np.uint8,
+        bg: np.ndarray,          # float32 (H, W) - scena intera
+        y0: nb.int64, x0: nb.int64,
+        wh: nb.int64, ww: nb.int64,
+    ) -> np.ndarray:
+        """Score rescored valutato SOLO nella finestra (y0, x0, wh, ww).
+
+        Le feature campionano lo spread dell'intera scena (bounds-checked
+        sui bordi scena): a differenza di chiamare il kernel su un crop,
+        le feature che escono dalla finestra NON contano come miss.
+        Usato dal path pyramid_propagate: costo ∝ area finestra.
+        """
+        H, W = spread.shape
+        N = dx.shape[0]
+        acc = np.zeros((wh, ww), dtype=np.float32)
+        for yi in nb.prange(wh):
+            y = y0 + yi
+            for i in range(N):
+                b = bins[i]
+                mask = np.uint8(1) << b
+                if (bit_active & mask) == 0:
+                    continue
+                yy = y + dy[i]
+                if yy < 0 or yy >= H:
+                    continue
+                ddx = dx[i]
+                xi_lo = 0
+                xi_hi = ww
+                lo = -(x0 + ddx)
+                if lo > xi_lo:
+                    xi_lo = lo
+                hi = W - (x0 + ddx)
+                if hi < xi_hi:
+                    xi_hi = hi
+                for xi in range(xi_lo, xi_hi):
+                    if spread[yy, x0 + xi + ddx] & mask:
+                        acc[yi, xi] += 1.0
+        if N > 0:
+            inv = 1.0 / N
+            for yi in nb.prange(wh):
+                for xi in range(ww):
+                    v = acc[yi, xi] * inv
+                    bgv = bg[y0 + yi, x0 + xi]
+                    if bgv < 1.0:
+                        r = (v - bgv) / (1.0 - bgv + 1e-6)
+                        acc[yi, xi] = r if r > 0.0 else 0.0
+                    else:
+                        acc[yi, xi] = 0.0
+        return acc
+
+    @nb.njit(cache=True, parallel=True, fastmath=True, boundscheck=False)
+    def _jit_score_bitmap_rescored_window_u16(
+        spread: np.ndarray,      # uint16 (H, W)
+        dx: np.ndarray, dy: np.ndarray, bins: np.ndarray,
+        bit_active: np.uint16,
+        bg: np.ndarray,
+        y0: nb.int64, x0: nb.int64,
+        wh: nb.int64, ww: nb.int64,
+    ) -> np.ndarray:
+        """Versione uint16 (polarity 16-bin) del kernel windowed."""
+        H, W = spread.shape
+        N = dx.shape[0]
+        acc = np.zeros((wh, ww), dtype=np.float32)
+        for yi in nb.prange(wh):
+            y = y0 + yi
+            for i in range(N):
+                b = bins[i]
+                mask = np.uint16(1) << b
+                if (bit_active & mask) == 0:
+                    continue
+                yy = y + dy[i]
+                if yy < 0 or yy >= H:
+                    continue
+                ddx = dx[i]
+                xi_lo = 0
+                xi_hi = ww
+                lo = -(x0 + ddx)
+                if lo > xi_lo:
+                    xi_lo = lo
+                hi = W - (x0 + ddx)
+                if hi < xi_hi:
+                    xi_hi = hi
+                for xi in range(xi_lo, xi_hi):
+                    if spread[yy, x0 + xi + ddx] & mask:
+                        acc[yi, xi] += 1.0
+        if N > 0:
+            inv = 1.0 / N
+            for yi in nb.prange(wh):
+                for xi in range(ww):
+                    v = acc[yi, xi] * inv
+                    bgv = bg[y0 + yi, x0 + xi]
+                    if bgv < 1.0:
+                        r = (v - bgv) / (1.0 - bgv + 1e-6)
+                        acc[yi, xi] = r if r > 0.0 else 0.0
+                    else:
+                        acc[yi, xi] = 0.0
+        return acc
+
     @nb.njit(cache=True, parallel=True, fastmath=True, boundscheck=False)
     def _jit_top_max_per_variant(
         spread: np.ndarray,            # uint8 (H, W)
@@ -426,6 +528,9 @@ if HAS_NUMBA:
         _jit_top_max_per_variant(
             spread, dx, dy, b, offsets, np.uint8(0xFF), bg_pv, scale_idx,
         )
+        _jit_score_bitmap_rescored_window(
+            spread, dx, dy, b, np.uint8(0xFF), bg, 4, 4, 8, 8,
+        )
         _jit_popcount_density(spread)
         spread16 = np.zeros((32, 32), dtype=np.uint16)
         _jit_score_bitmap_rescored_u16(
@@ -447,6 +552,12 @@ else:  # pragma: no cover
     def _jit_score_bitmap_rescored_strided(spread, dx, dy, bins, bit_active, bg, stride):
         raise RuntimeError("numba non disponibile")
 
+    def _jit_score_bitmap_rescored_window(spread, dx, dy, bins, bit_active, bg, y0, x0, wh, ww):
+        raise RuntimeError("numba non disponibile")
+
+    def _jit_score_bitmap_rescored_window_u16(spread, dx, dy, bins, bit_active, bg, y0, x0, wh, ww):
+        raise RuntimeError("numba non disponibile")
+
     def _jit_score_bitmap_greedy(spread, dx, dy, bins, bit_active, min_score, greediness):
         raise RuntimeError("numba non disponibile")
 
@@ -524,6 +635,39 @@ def score_bitmap_rescored(
     return np.maximum(0.0, out).astype(np.float32)
 
 
+def score_bitmap_rescored_window(
+    spread: np.ndarray, dx: np.ndarray, dy: np.ndarray, bins: np.ndarray,
+    bit_active: int, bg: np.ndarray,
+    y0: int, x0: int, wh: int, ww: int,
+) -> np.ndarray:
+    """Score rescored solo nella finestra (y0, x0, wh, ww) della scena.
+
+    Le feature campionano l'INTERA scena: feature fuori finestra ma dentro
+    scena contano correttamente (chiamare il kernel su un crop le tratta
+    come miss e azzera lo score — il bug che rendeva inutilizzabile il
+    path pyramid_propagate). Fallback no-numba: kernel pieno + slice.
+    """
+    if HAS_NUMBA and len(dx) > 0:
+        dx_c = np.ascontiguousarray(dx, dtype=np.int32)
+        dy_c = np.ascontiguousarray(dy, dtype=np.int32)
+        bins_c = np.ascontiguousarray(bins, dtype=np.int8)
+        bg_c = np.ascontiguousarray(bg, dtype=np.float32)
+        if spread.dtype == np.uint16:
+            return _jit_score_bitmap_rescored_window_u16(
+                np.ascontiguousarray(spread, dtype=np.uint16),
+                dx_c, dy_c, bins_c, np.uint16(bit_active), bg_c,
+                int(y0), int(x0), int(wh), int(ww),
+            )
+        return _jit_score_bitmap_rescored_window(
+            np.ascontiguousarray(spread, dtype=np.uint8),
+            dx_c, dy_c, bins_c, np.uint8(bit_active), bg_c,
+            int(y0), int(x0), int(wh), int(ww),
+        )
+    # Fallback (lento, solo senza numba): score full-frame + slice finestra
+    full = score_bitmap_rescored(spread, dx, dy, bins, bit_active, bg)
+    return full[y0:y0 + wh, x0:x0 + ww]
+
+
 def score_bitmap_greedy(
     spread: np.ndarray, dx: np.ndarray, dy: np.ndarray, bins: np.ndarray,
     bit_active: int, min_score: float, greediness: float,

pm2d/auto_tune.py

@@ -61,6 +61,8 @@ def detect_rotational_symmetry(
 
     center = (w / 2.0, h / 2.0)
     ref = mag
+    # ref è costante nel loop sugli angoli: centra una volta sola
+    rm = ref - ref.mean()
 
     correlations: list[tuple[float, float]] = []
     for ang in np.arange(step_deg, 360.0, step_deg):
@@ -68,7 +70,6 @@ def detect_rotational_symmetry(
         rot = cv2.warpAffine(
             mag, M, (w, h), borderValue=0.0,
         )
-        rm = ref - ref.mean()
         rs = rot - rot.mean()
         denom = np.sqrt((rm * rm).sum() * (rs * rs).sum()) + 1e-9
         c = float((rm * rs).sum() / denom)

pm2d/dxf.py

@@ -0,0 +1,119 @@
+"""Rasterizzazione DXF → immagine template per il matcher shape-based.
+
+Il matcher lavora sui gradienti degli edge: un line-drawing pulito
+(sfondo grigio scuro, tratti chiari) è un template perfettamente valido.
+Questo modulo converte un file DXF (CAD 2D) in una bitmap grayscale
+centrata e scalata, pronta per train().
+"""
+from __future__ import annotations
+
+import io
+
+import cv2
+import numpy as np
+
+# Valori di rendering: sfondo scuro / tratto chiaro → gradiente netto
+BG_GRAY = 60
+LINE_GRAY = 220
+
+
+def _read_doc(data: bytes):
+    """Parse DXF da bytes con gestione encoding.
+
+    Prima prova ezdxf.read su StringIO (DXF ASCII utf-8 / cp1252),
+    poi fallback su ezdxf.recover che auto-rileva encoding e tollera
+    file malformati.
+    """
+    import ezdxf
+    from ezdxf import recover
+
+    for enc in ("utf-8", "cp1252"):
+        try:
+            text = data.decode(enc)
+            return ezdxf.read(io.StringIO(text))
+        except Exception:
+            # UnicodeDecodeError, DXFStructureError e simili: prossimo tentativo
+            continue
+    # Ultimo tentativo: recover lavora direttamente sui bytes
+    try:
+        doc, _auditor = recover.read(io.BytesIO(data))
+        return doc
+    except Exception as e:
+        raise ValueError(f"DXF illeggibile o corrotto: {e}") from e
+
+
+def _extract_polylines(doc, flatten_dist: float = 0.05) -> tuple[list[np.ndarray], int]:
+    """Converte le entità del modelspace in polilinee (liste di punti XY).
+
+    Entità non convertibili (non supportate da make_path) vengono saltate
+    silenziosamente ma conteggiate. Ritorna (polilinee, n_saltate).
+    """
+    from ezdxf import path as ezpath
+
+    polylines: list[np.ndarray] = []
+    skipped = 0
+    for entity in doc.modelspace():
+        try:
+            p = ezpath.make_path(entity)
+            pts = np.array(
+                [(v.x, v.y) for v in p.flattening(distance=flatten_dist)],
+                dtype=np.float64,
+            )
+            if len(pts) >= 2:
+                polylines.append(pts)
+        except Exception:
+            skipped += 1
+    return polylines, skipped
+
+
+def dxf_to_image(data: bytes, target_size: int = 512,
+                 line_thickness: int = 2, margin: int = 16) -> np.ndarray:
+    """Rasterizza un DXF in immagine grayscale (H, W) uint8.
+
+    - Scala uniforme: il lato lungo del disegno = target_size - 2*margin.
+    - Disegno centrato, asse Y CAD (su) ribaltato in convenzione immagine.
+    - Sfondo grigio scuro (60), tratti chiari (220), antialiased.
+
+    Solleva ValueError se il DXF è vuoto o illeggibile.
+    """
+    doc = _read_doc(data)
+
+    # Distanza di flattening provvisoria in unità CAD: raffinata sotto
+    # una volta nota la scala (qui serve solo per il bounding box).
+    polylines, skipped = _extract_polylines(doc)
+    if not polylines:
+        raise ValueError(
+            "DXF vuoto: nessuna entità convertibile in polilinea nel "
+            f"modelspace ({skipped} entità non supportate saltate)")
+
+    all_pts = np.vstack(polylines)
+    min_xy = all_pts.min(axis=0)
+    max_xy = all_pts.max(axis=0)
+    extent = max_xy - min_xy
+    long_side = float(extent.max())
+    if long_side <= 0:
+        raise ValueError("DXF degenere: bounding box con estensione nulla")
+
+    # Ri-flattening con distanza adattiva: ~0.25 px di errore alla scala
+    # finale (il primo pass usava una tolleranza in unità CAD arbitraria).
+    avail = max(1, target_size - 2 * margin)
+    scale = avail / long_side
+    polylines, _ = _extract_polylines(doc, flatten_dist=max(1e-9, 0.25 / scale))
+
+    canvas = np.full((target_size, target_size), BG_GRAY, dtype=np.uint8)
+    # Offset per centrare il disegno (anche sul lato corto)
+    draw_w = extent[0] * scale
+    draw_h = extent[1] * scale
+    off_x = (target_size - draw_w) / 2.0
+    off_y = (target_size - draw_h) / 2.0
+
+    for pts in polylines:
+        px = (pts[:, 0] - min_xy[0]) * scale + off_x
+        # Y CAD verso l'alto → Y immagine verso il basso
+        py = (max_xy[1] - pts[:, 1]) * scale + off_y
+        ipts = np.stack([px, py], axis=1).round().astype(np.int32)
+        cv2.polylines(canvas, [ipts], isClosed=False,
+                      color=LINE_GRAY, thickness=line_thickness,
+                      lineType=cv2.LINE_AA)
+
+    return canvas

pm2d/gui.py

@@ -12,7 +12,6 @@ Tutta la logica algoritmica vive in pm2d.matcher.EdgeShapeMatcher.
 
 from __future__ import annotations
 
-import sys
 from pathlib import Path
 from tkinter import Tk, filedialog
 import tkinter as tk
@@ -196,8 +195,10 @@ def _warp_template_edges_to_scene(
     edge = cv2.Canny(template_gray, canny_low, canny_high)
     # Matrice affine: scala + rotazione attorno al centro template, poi traslazione
     Ht, Wt = h, w
-    cx_t = (Wt - 1) / 2.0
-    cy_t = (Ht - 1) / 2.0
+    # Centro coerente con la convenzione train (center = w / 2.0, no -1):
+    # (Wt-1)/2 introduceva uno shift di 0.5px per template di lato pari.
+    cx_t = Wt / 2.0
+    cy_t = Ht / 2.0
     M = cv2.getRotationMatrix2D((cx_t, cy_t), angle_deg, scale)
     # Traslazione per portare centro template a (cx, cy) della scena
     M[0, 2] += cx - cx_t
@@ -492,7 +493,9 @@ def run(
     num_features: int = 96,
     weak_grad: float = 30.0,
     strong_grad: float = 60.0,
-    spread_radius: int = 5,
+    # 4 allineato col default del matcher: raggio 5 peggiora la precisione
+    # di rotazione (spread troppo largo appiattisce il picco angolare).
+    spread_radius: int = 4,
     pyramid_levels: int = 3,
     min_score: float = 0.55,
     max_matches: int = 25,

pm2d/line_matcher.py

@@ -38,8 +38,8 @@ _GOLDEN = (math.sqrt(5.0) - 1.0) / 2.0  # ≈ 0.618
 
 from pm2d._jit_kernels import (
     score_by_shift as _jit_score_by_shift,
-    score_bitmap as _jit_score_bitmap,
     score_bitmap_rescored as _jit_score_bitmap_rescored,
+    score_bitmap_rescored_window as _jit_score_bitmap_rescored_window,
     score_bitmap_greedy as _jit_score_bitmap_greedy,
     top_max_per_variant as _jit_top_max_per_variant,
     popcount_density as _jit_popcount,
@@ -172,7 +172,7 @@ class LineShapeMatcher:
         scale_step: float = 0.1,
         spread_radius: int = 4,
         min_feature_spacing: int = 3,
-        pyramid_levels: int = 2,
+        pyramid_levels: int = 3,
         top_score_factor: float = 0.5,
         n_threads: int | None = None,
         use_polarity: bool = False,
@@ -325,8 +325,6 @@ class LineShapeMatcher:
         n_vars = len(self.variants)
         n_levels = len(self.variants[0].levels)
         var_meta = np.zeros((n_vars, 6), dtype=np.float32)  # ang, scale, kh, kw, cxl, cyl
-        all_dx, all_dy, all_bin, all_offsets = [], [], [], []
-        offset = 0
         all_offsets_per_level = [[] for _ in range(n_levels)]
         all_dx_per_level = [[] for _ in range(n_levels)]
         all_dy_per_level = [[] for _ in range(n_levels)]
@@ -473,8 +471,46 @@ class LineShapeMatcher:
         step = self._effective_angle_step()
         if step <= 0 or a0 >= a1:
             return [float(a0)]
-        n = int(np.floor((a1 - a0) / step))
-        return [float(a0 + i * step) for i in range(n)]
+        # Include l'estremo superiore: con range parziali (es. ±15°) il
+        # +15° deve essere testato quanto il -15°. Se il range copre 360°
+        # interi l'estremo coincide con a0 (mod 360) e viene escluso per
+        # non duplicare la variante.
+        n = int(np.floor((a1 - a0) / step + 1e-9)) + 1
+        angles = [float(a0 + i * step) for i in range(n)]
+        if a1 - a0 >= 360.0:
+            angles = [a for a in angles if a - a0 < 360.0 - 1e-9]
+        return angles
+
+    def _prepare_padded_template(
+        self, template_gray: np.ndarray, mask_full: np.ndarray, scale: float,
+    ) -> tuple[np.ndarray, np.ndarray, tuple[float, float], int]:
+        """Scala + padda template e mask; ritorna (gray_p, mask_p, center, diag).
+
+        `center` e' il centro REALE del template dentro l'immagine paddata
+        (px + sw/2, py + sh/2): con padding floor differisce da diag/2 fino
+        a 0.5 px. Ruotare attorno a diag/2 (come si faceva prima) faceva
+        orbitare il centro-modello attorno al centro di rotazione, con un
+        bias di posizione dipendente dall'angolo. Tutti i percorsi che
+        ricostruiscono il template ruotato devono usare QUESTO helper.
+        """
+        h, w = template_gray.shape
+        sw = max(16, int(round(w * scale)))
+        sh = max(16, int(round(h * scale)))
+        gray_s = cv2.resize(template_gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
+        mask_s = cv2.resize(mask_full, (sw, sh), interpolation=cv2.INTER_NEAREST)
+        diag = int(np.ceil(np.hypot(sh, sw))) + 6
+        py = (diag - sh) // 2
+        px = (diag - sw) // 2
+        gray_p = cv2.copyMakeBorder(
+            gray_s, py, diag - sh - py, px, diag - sw - px,
+            cv2.BORDER_REPLICATE,
+        )
+        mask_p = cv2.copyMakeBorder(
+            mask_s, py, diag - sh - py, px, diag - sw - px,
+            cv2.BORDER_CONSTANT, value=0,
+        )
+        center = (px + sw / 2.0, py + sh / 2.0)
+        return gray_p, mask_p, center, diag
 
     # --- Training ------------------------------------------------------
 
@@ -504,6 +540,13 @@ class LineShapeMatcher:
         h, w = gray.shape
         self.template_size = (w, h)
         self.template_gray = gray.copy()
+        # Clamp livelli piramide alla dimensione template: al top-level il
+        # lato minimo deve restare >= 12 px, sotto le feature collassano
+        # tutte negli stessi (dx,dy) e lo score top diventa rumore.
+        max_lv = 1
+        while min(w, h) / (2 ** max_lv) >= 12 and max_lv < 4:
+            max_lv += 1
+        self.pyramid_levels = max(1, min(self.pyramid_levels, max_lv))
         if mask is None:
             mask_full = np.full((h, w), 255, dtype=np.uint8)
         else:
@@ -566,24 +609,10 @@ class LineShapeMatcher:
         Estrazione algorithm identica al train() originale, separato per
         riuso da add_template_view (multi-template ensemble).
         """
-        h, w = gray.shape
         for s in self._scale_list():
-            sw = max(16, int(round(w * s)))
-            sh = max(16, int(round(h * s)))
-            gray_s = cv2.resize(gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
-            mask_s = cv2.resize(mask_full, (sw, sh), interpolation=cv2.INTER_NEAREST)
-            diag = int(np.ceil(np.hypot(sh, sw))) + 6
-            py = (diag - sh) // 2
-            px = (diag - sw) // 2
-            gray_p = cv2.copyMakeBorder(
-                gray_s, py, diag - sh - py, px, diag - sw - px,
-                cv2.BORDER_REPLICATE,
+            gray_p, mask_p, center, diag = self._prepare_padded_template(
+                gray, mask_full, s,
             )
-            mask_p = cv2.copyMakeBorder(
-                mask_s, py, diag - sh - py, px, diag - sw - px,
-                cv2.BORDER_CONSTANT, value=0,
-            )
-            center = (diag / 2.0, diag / 2.0)
 
             for ang in self._angle_list():
                 M = cv2.getRotationMatrix2D(center, ang, 1.0)
@@ -600,10 +629,10 @@ class LineShapeMatcher:
                 if len(fx) < 8:
                     continue
 
-                cx_c = diag / 2.0
-                cy_c = diag / 2.0
-                dx = (fx - cx_c).astype(np.int32)
-                dy = (fy - cy_c).astype(np.int32)
+                # round (non truncation): astype(int32) tronca verso zero
+                # e introduceva un bias sistematico ~0.25 px verso il centro.
+                dx = np.round(fx - center[0]).astype(np.int32)
+                dy = np.round(fy - center[1]).astype(np.int32)
 
                 x0 = int(dx.min()); x1 = int(dx.max())
                 y0 = int(dy.min()); y1 = int(dy.max())
@@ -687,8 +716,13 @@ class LineShapeMatcher:
         try:
             import hashlib
             h = hashlib.md5()
-            sample = gray.tobytes()[:65536]
-            h.update(sample)
+            # Hash dell'INTERA scena: hashare solo i primi 64KB (prime
+            # ~80 righe a 830px) faceva collidere scene con la stessa
+            # banda superiore (es. sfondo uniforme da camera fissa) →
+            # find() ritornava i risultati della scena sbagliata.
+            # tobytes() copiava gia' tutto il buffer, il costo extra
+            # dell'md5 completo e' ~1ms.
+            h.update(gray.tobytes())
             h.update(f"|{gray.shape}|{gray.dtype}".encode())
             h.update(
                 f"|{self.weak_grad}|{self.strong_grad}"
@@ -717,11 +751,17 @@ class LineShapeMatcher:
         while len(self._scene_cache) > self._SCENE_CACHE_SIZE:
             self._scene_cache.popitem(last=False)
 
-    def _spread_bitmap(self, gray: np.ndarray) -> np.ndarray:
+    def _spread_bitmap(
+        self, gray: np.ndarray, radius: int | None = None,
+    ) -> np.ndarray:
         """Spread bitmap: bit b acceso dove bin b è presente nel raggio.
 
         dtype: uint8 per N_BINS=8, uint16 per N_BINS_POL=16 (use_polarity).
         Se use_gpu=True: Sobel + dilate via cv2.UMat (OpenCL kernel GPU).
+        radius: override del raggio di spread (default self.spread_radius).
+        radius=0/1 produce una bitmap "fine" senza tolleranza, usata nel
+        refine finale: sulla bitmap dilatata lo score satura e il refine
+        non distingue pose entro ±spread_radius px / ±atan(spread/R) gradi.
         """
         if self.use_gpu and not isinstance(gray, cv2.UMat):
             gray_in = cv2.UMat(np.ascontiguousarray(gray))
@@ -729,7 +769,8 @@ class LineShapeMatcher:
             gray_in = gray
         mag, bins = self._gradient(gray_in)
         valid = mag >= self.weak_grad
-        k = 2 * self.spread_radius + 1
+        r = self.spread_radius if radius is None else max(0, int(radius))
+        k = 2 * r + 1
         kernel = np.ones((k, k), dtype=np.uint8)
         H, W = (gray.shape if isinstance(gray, np.ndarray)
                 else (gray.get().shape[0], gray.get().shape[1]))
@@ -755,7 +796,9 @@ class LineShapeMatcher:
             if not bin_present[b]:
                 continue  # XX: nessun pixel di questo bin sopra weak_grad
             mask_b = ((bins == b) & valid).astype(np.uint8)
-            if self.use_gpu:
+            if r == 0:
+                d_np = mask_b
+            elif self.use_gpu:
                 d = cv2.dilate(cv2.UMat(mask_b), kernel)
                 d_np = d.get()
             else:
@@ -828,111 +871,9 @@ class LineShapeMatcher:
         oy = float(np.clip(oy, -0.5, 0.5))
         return x + ox, y + oy
 
-    def _refine_pose_joint(
-        self,
-        spread0: np.ndarray,
-        template_gray: np.ndarray,
-        cx: float, cy: float,
-        angle_deg: float, scale: float,
-        mask_full: np.ndarray,
-        max_iter: int = 24,
-        tol: float = 1e-3,
-    ) -> tuple[float, float, float, float]:
-        """Refine congiunto (cx, cy, angle) via Nelder-Mead 3D.
-
-        Ottimizza simultaneamente posizione e angolo (vs golden search 1D
-        sull'angolo poi quadratico 2D su xy che alterna assi). Halcon-style:
-        un singolo iter LM stila il match a precisione sub-pixel + sub-step.
-        Ritorna (angle, score, cx, cy) dove score e quello calcolato sulla
-        scena spread (no template gray).
-        """
-        h, w = template_gray.shape
-        sw = max(16, int(round(w * scale)))
-        sh = max(16, int(round(h * scale)))
-        gray_s = cv2.resize(template_gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
-        mask_s = cv2.resize(mask_full, (sw, sh), interpolation=cv2.INTER_NEAREST)
-        diag = int(np.ceil(np.hypot(sh, sw))) + 6
-        py = (diag - sh) // 2; px = (diag - sw) // 2
-        gray_p = cv2.copyMakeBorder(gray_s, py, diag - sh - py, px, diag - sw - px,
-                                     cv2.BORDER_REPLICATE)
-        mask_p = cv2.copyMakeBorder(mask_s, py, diag - sh - py, px, diag - sw - px,
-                                     cv2.BORDER_CONSTANT, value=0)
-        center = (diag / 2.0, diag / 2.0)
-        H, W = spread0.shape
-
-        def _score(params: tuple[float, float, float]) -> float:
-            ddx, ddy, dang = params
-            ang = angle_deg + dang
-            M = cv2.getRotationMatrix2D(center, ang, 1.0)
-            gray_r = cv2.warpAffine(gray_p, M, (diag, diag),
-                                     flags=cv2.INTER_LINEAR,
-                                     borderMode=cv2.BORDER_REPLICATE)
-            mask_r = cv2.warpAffine(mask_p, M, (diag, diag),
-                                     flags=cv2.INTER_NEAREST, borderValue=0)
-            mag, bins = self._gradient(gray_r)
-            fx, fy, fb = self._extract_features(mag, bins, mask_r)
-            if len(fx) < 8:
-                return 0.0
-            cxe = cx + ddx; cye = cy + ddy
-            ix = int(round(cxe)); iy = int(round(cye))
-            tot = 0
-            valid = 0
-            for i in range(len(fx)):
-                xs = ix + int(fx[i] - center[0])
-                ys = iy + int(fy[i] - center[1])
-                if 0 <= xs < W and 0 <= ys < H:
-                    bit = np.uint8(1 << int(fb[i]))
-                    if spread0[ys, xs] & bit:
-                        tot += 1
-                    valid += 1
-            return -float(tot) / max(1, valid)  # minimize -score
-
-        # Nelder-Mead 3D inline (no scipy). Simplex iniziale: vertice + offset
-        # dx=±0.5px, dy=±0.5px, dθ=±step/2.
-        step_a = self.angle_step_deg / 2.0 if self.angle_step_deg > 0 else 1.0
-        x0 = np.array([0.0, 0.0, 0.0])
-        simplex = np.array([
-            x0,
-            x0 + [0.5, 0.0, 0.0],
-            x0 + [0.0, 0.5, 0.0],
-            x0 + [0.0, 0.0, step_a],
-        ])
-        fvals = np.array([_score(tuple(s)) for s in simplex])
-        for _ in range(max_iter):
-            order = np.argsort(fvals)
-            simplex = simplex[order]; fvals = fvals[order]
-            if abs(fvals[-1] - fvals[0]) < tol:
-                break
-            centroid = simplex[:-1].mean(axis=0)
-            xr = centroid + 1.0 * (centroid - simplex[-1])
-            fr = _score(tuple(xr))
-            if fvals[0] <= fr < fvals[-2]:
-                simplex[-1] = xr; fvals[-1] = fr
-                continue
-            if fr < fvals[0]:
-                xe = centroid + 2.0 * (centroid - simplex[-1])
-                fe = _score(tuple(xe))
-                if fe < fr:
-                    simplex[-1] = xe; fvals[-1] = fe
-                else:
-                    simplex[-1] = xr; fvals[-1] = fr
-                continue
-            xc = centroid + 0.5 * (simplex[-1] - centroid)
-            fc = _score(tuple(xc))
-            if fc < fvals[-1]:
-                simplex[-1] = xc; fvals[-1] = fc
-                continue
-            for k in range(1, 4):
-                simplex[k] = simplex[0] + 0.5 * (simplex[k] - simplex[0])
-                fvals[k] = _score(tuple(simplex[k]))
-        best_i = int(np.argmin(fvals))
-        ddx, ddy, dang = simplex[best_i]
-        return (angle_deg + float(dang), -float(fvals[best_i]),
-                cx + float(ddx), cy + float(ddy))
-
     def _refine_angle(
         self,
-        spread0: np.ndarray,       # bitmap uint8 (H, W)
+        spread0: np.ndarray,       # bitmap uint8/uint16 (H, W) - spread pieno
         bit_active: int,
         template_gray: np.ndarray,
         cx: float, cy: float,
@@ -941,33 +882,31 @@ class LineShapeMatcher:
         angle_fine_step: float = 0.5,
         search_radius: float | None = None,
         original_score: float | None = None,
+        spread_fine: np.ndarray | None = None,
     ) -> tuple[float, float, float, float]:
         """Ricerca angolare fine (sub-step) attorno al match grezzo.
 
-        Genera 5 template temporanei a angle ± {0.5, 1.0} * step e sceglie
-        l'angolo con score massimo (parabolic fit sulle 3 score centrali).
+        Golden-section sull'angolo + argmax posizione in finestra ±3 px.
         Ritorna (angle_refined, score, cx_refined, cy_refined).
+
+        L'ottimizzazione gira sulla bitmap FINE (spread_fine, raggio 1):
+        sulla bitmap dilatata (spread0, raggio 4-5) lo score satura a 1.0
+        per qualunque posa entro ±spread px / ±atan(spread/R) gradi e il
+        refine non vede alcun gradiente (l'angolo restava quello grezzo
+        quantizzato e cv2.minMaxLoc sul plateau saturo spostava il centro
+        sull'angolo in alto a sinistra della finestra: errore misurato
+        3·sqrt(2) ≈ 4.24 px). Lo score RITORNATO e' ricalcolato alla posa
+        raffinata su spread0, per mantenere la semantica precedente
+        (tolleranza spread_radius) su soglie/min_score.
         """
-        # NB: rimosso early-skip su score >= 0.99. Lo score linemod/shape
-        # satura facilmente a 1.0 (specie con pyramid_propagate o spread
-        # ampio) ma NON garantisce angolo preciso: l'angolo grezzo della
-        # variante e' quantizzato a multipli di angle_step (5 deg default).
-        # Refine angolare e' essenziale per orientamento sub-step.
         if search_radius is None:
             search_radius = self._effective_angle_step()
+        # Bitmap su cui ottimizzare: fine se disponibile, altrimenti spread0.
+        opt_map = spread_fine if spread_fine is not None else spread0
 
-        h, w = template_gray.shape
-        sw = max(16, int(round(w * scale)))
-        sh = max(16, int(round(h * scale)))
-        gray_s = cv2.resize(template_gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
-        mask_s = cv2.resize(mask_full, (sw, sh), interpolation=cv2.INTER_NEAREST)
-        diag = int(np.ceil(np.hypot(sh, sw))) + 6
-        py = (diag - sh) // 2; px = (diag - sw) // 2
-        gray_p = cv2.copyMakeBorder(gray_s, py, diag - sh - py, px, diag - sw - px,
-                                     cv2.BORDER_REPLICATE)
-        mask_p = cv2.copyMakeBorder(mask_s, py, diag - sh - py, px, diag - sw - px,
-                                     cv2.BORDER_CONSTANT, value=0)
-        center = (diag / 2.0, diag / 2.0)
+        gray_p, mask_p, center, diag = self._prepare_padded_template(
+            template_gray, mask_full, scale,
+        )
 
         H, W = spread0.shape
         margin = 3
@@ -982,14 +921,11 @@ class LineShapeMatcher:
         feat_cache = self._refine_feat_cache
         cache_scale_key = round(scale * 1000)
 
-        def _score_at_angle(off: float) -> tuple[float, float, float]:
-            """Ritorna (score, best_cx, best_cy) per angolo = angle_deg + off."""
-            ang = angle_deg + off
+        def _feats_at_angle(ang: float):
             ck = (round(ang * 20), cache_scale_key)
             cached = feat_cache.get(ck)
             if cached is not None:
-                fx, fy, fb = cached
-            else:
+                return cached
             M = cv2.getRotationMatrix2D(center, ang, 1.0)
             gray_r = cv2.warpAffine(gray_p, M, (diag, diag),
                                      flags=cv2.INTER_LINEAR,
@@ -1002,15 +938,27 @@ class LineShapeMatcher:
             if len(feat_cache) > 256:
                 feat_cache.pop(next(iter(feat_cache)))
             feat_cache[ck] = (fx, fy, fb)
+            return fx, fy, fb
+
+        def _score_at_angle(off: float) -> tuple[float, float, float]:
+            """Ritorna (score, best_cx, best_cy) per angolo = angle_deg + off.
+
+            Score = max su finestra ±margin px attorno a (cx, cy) sulla
+            bitmap di ottimizzazione; posizione = picco sub-pixel della
+            finestra (centroide plateau / fit quadratico, NON minMaxLoc
+            che sul plateau e' biased verso l'angolo top-left).
+            """
+            ang = angle_deg + off
+            fx, fy, fb = _feats_at_angle(ang)
             if len(fx) < 8:
                 return (0.0, cx, cy)
-            dx = (fx - center[0]).astype(np.int32)
-            dy = (fy - center[1]).astype(np.int32)
+            dx = np.round(fx - center[0]).astype(np.int32)
+            dy = np.round(fy - center[1]).astype(np.int32)
             y_lo = int(cy) - margin; y_hi = int(cy) + margin + 1
             x_lo = int(cx) - margin; x_hi = int(cx) + margin + 1
             sh_w = y_hi - y_lo; sw_w = x_hi - x_lo
             acc = np.zeros((sh_w, sw_w), dtype=np.float32)
-            spread_dtype = spread0.dtype.type
+            spread_dtype = opt_map.dtype.type
             for i in range(len(dx)):
                 ddx = int(dx[i]); ddy = int(dy[i]); b = int(fb[i])
                 bit = spread_dtype(1 << b)
@@ -1021,14 +969,19 @@ class LineShapeMatcher:
                 s_y0 = max(0, sy0); s_y1 = min(H, sy1)
                 s_x0 = max(0, sx0); s_x1 = min(W, sx1)
                 if s_y1 > s_y0 and s_x1 > s_x0:
-                    region = spread0[s_y0:s_y1, s_x0:s_x1]
+                    region = opt_map[s_y0:s_y1, s_x0:s_x1]
                     acc[a_y0:a_y1, a_x0:a_x1] += (
                         (region & bit) != 0
                     ).astype(np.float32)
             acc /= len(dx)
             _, max_val, _, max_loc = cv2.minMaxLoc(acc)
-            return (float(max_val),
-                    float(x_lo + max_loc[0]), float(y_lo + max_loc[1]))
+            if max_val <= 0.0:
+                return (0.0, cx, cy)
+            # Picco sub-pixel dentro la finestra (gestisce plateau e fit 3x3)
+            px_f, py_f = self._subpixel_peak(
+                acc, int(max_loc[0]), int(max_loc[1]), plateau_radius=margin,
+            )
+            return (float(max_val), float(x_lo + px_f), float(y_lo + py_f))
 
         # Golden-section search su [-search_radius, +search_radius]:
         # converge in log tempo a precisione ~0.1°, ~8 valutazioni vs 5
@@ -1064,7 +1017,25 @@ class LineShapeMatcher:
                 x1 = x2; s1 = s2; cx1 = cx2; cy1 = cy2
                 x2 = a_lo + _GOLDEN * (a_hi - a_lo)
                 s2, cx2, cy2 = _score_at_angle(x2)
+
+        ang_best, s_best, cx_best, cy_best = best
+        if spread_fine is None:
             return best
+        # Score finale alla posa raffinata sullo spread COARSE: stessa
+        # semantica dello score pre-refine (tolleranza spread_radius),
+        # cosi' min_score/verify mantengono il significato di prima.
+        fx, fy, fb = _feats_at_angle(ang_best)
+        if len(fx) < 8:
+            return best
+        xs = np.round(fx - center[0]).astype(np.int32) + int(round(cx_best))
+        ys = np.round(fy - center[1]).astype(np.int32) + int(round(cy_best))
+        ok = (xs >= 0) & (xs < W) & (ys >= 0) & (ys < H)
+        if not ok.any():
+            return (ang_best, 0.0, cx_best, cy_best)
+        bits = spread0[ys[ok], xs[ok]].astype(np.int32)
+        hit = (bits & np.left_shift(1, fb[ok].astype(np.int32))) != 0
+        score_coarse = float(hit.sum()) / len(fx)
+        return (ang_best, score_coarse, cx_best, cy_best)
 
     def _get_view_template(
         self, view_idx: int,
@@ -1089,26 +1060,13 @@ class LineShapeMatcher:
         """
         if self.template_gray is None:
             return 1.0
-        h, w = self.template_gray.shape
-        scale = variant.scale
-        sw = max(16, int(round(w * scale)))
-        sh = max(16, int(round(h * scale)))
-        gray_s = cv2.resize(self.template_gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
         mask_src = (
             self._train_mask if self._train_mask is not None
             else np.full_like(self.template_gray, 255)
         )
-        mask_s = cv2.resize(mask_src, (sw, sh), interpolation=cv2.INTER_NEAREST)
-        diag = int(np.ceil(np.hypot(sh, sw))) + 6
-        py = (diag - sh) // 2; px = (diag - sw) // 2
-        gray_p = cv2.copyMakeBorder(
-            gray_s, py, diag - sh - py, px, diag - sw - px, cv2.BORDER_REPLICATE,
+        gray_p, mask_p, center, diag = self._prepare_padded_template(
+            self.template_gray, mask_src, variant.scale,
         )
-        mask_p = cv2.copyMakeBorder(
-            mask_s, py, diag - sh - py, px, diag - sw - px,
-            cv2.BORDER_CONSTANT, value=0,
-        )
-        center = (diag / 2.0, diag / 2.0)
         M = cv2.getRotationMatrix2D(center, angle_deg, 1.0)
         gray_r = cv2.warpAffine(gray_p, M, (diag, diag),
                                  flags=cv2.INTER_LINEAR,
@@ -1125,8 +1083,8 @@ class LineShapeMatcher:
         ix = int(round(cx)); iy = int(round(cy))
         hits = 0
         for i in range(n_feat):
-            xs = ix + int(fx[i] - center[0])
-            ys = iy + int(fy[i] - center[1])
+            xs = ix + int(round(fx[i] - center[0]))
+            ys = iy + int(round(fy[i] - center[1]))
             if 0 <= xs < W and 0 <= ys < H:
                 bit = spread_dtype(1 << int(fb[i]))
                 if spread0[ys, xs] & bit:
@@ -1140,26 +1098,13 @@ class LineShapeMatcher:
         """Soft-margin gradient similarity (Halcon Metric='use_polarity')."""
         if self.template_gray is None:
             return 0.0
-        h, w = self.template_gray.shape
-        scale = variant.scale
-        sw = max(16, int(round(w * scale)))
-        sh = max(16, int(round(h * scale)))
-        gray_s = cv2.resize(self.template_gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
         mask_src = (
             self._train_mask if self._train_mask is not None
             else np.full_like(self.template_gray, 255)
         )
-        mask_s = cv2.resize(mask_src, (sw, sh), interpolation=cv2.INTER_NEAREST)
-        diag = int(np.ceil(np.hypot(sh, sw))) + 6
-        py = (diag - sh) // 2; px = (diag - sw) // 2
-        gray_p = cv2.copyMakeBorder(
-            gray_s, py, diag - sh - py, px, diag - sw - px, cv2.BORDER_REPLICATE,
+        gray_p, mask_p, center, diag = self._prepare_padded_template(
+            self.template_gray, mask_src, variant.scale,
         )
-        mask_p = cv2.copyMakeBorder(
-            mask_s, py, diag - sh - py, px, diag - sw - px,
-            cv2.BORDER_CONSTANT, value=0,
-        )
-        center = (diag / 2.0, diag / 2.0)
         M = cv2.getRotationMatrix2D(center, angle_deg, 1.0)
         gray_r = cv2.warpAffine(gray_p, M, (diag, diag),
                                  flags=cv2.INTER_LINEAR,
@@ -1179,8 +1124,8 @@ class LineShapeMatcher:
         ix = int(round(cx)); iy = int(round(cy))
         sims = []; weights = []
         for i in range(len(fx)):
-            xs = ix + int(fx[i] - center[0])
-            ys = iy + int(fy[i] - center[1])
+            xs = ix + int(round(fx[i] - center[0]))
+            ys = iy + int(round(fy[i] - center[1]))
             if not (0 <= xs < W and 0 <= ys < H):
                 continue
             tx = float(gx_t[int(fy[i]), int(fx[i])])
@@ -1201,35 +1146,34 @@ class LineShapeMatcher:
     def _subpixel_refine_lm(
         self, scene_gray: np.ndarray, variant: _Variant,
         cx: float, cy: float, angle_deg: float,
-        n_iters: int = 2,
-    ) -> tuple[float, float]:
-        """Sub-pixel refinement iterativo via gradient-field least-squares.
+        n_iters: int = 4,
+        scene_grad: tuple[np.ndarray, np.ndarray] | None = None,
+    ) -> tuple[float, float, float]:
+        """Refinement least-squares congiunto di posizione E angolo.
 
-        Halcon-equivalent SubPixel='least_squares_high'. Precisione attesa
-        0.05 px (vs 0.5 px del fit quadratic 2D).
+        Halcon-equivalent SubPixel='least_squares_high'. Per ogni feature
+        template cerca il picco sub-pixel del gradiente scena lungo la
+        normale dell'edge (snap ±2 px, fit parabolico su 5 campioni), poi
+        risolve ai minimi quadrati pesati il sistema 3x3 in (dx, dy, dθ):
+
+            n_i · (d + dθ·u_i) = t_i,    u_i = (r_y,i, -r_x,i)
+
+        dove r_i = offset feature dal centro, n_i = normale edge template,
+        t_i = offset del picco lungo n_i, u_i = derivata della rotazione
+        nella convenzione cv2.getRotationMatrix2D (R = [[c,s],[-s,c]]).
+        Tra le iterazioni offset e normali vengono ruotati analiticamente
+        (no re-warp del template). Precisione attesa <0.1 px / <0.1°.
+
+        scene_grad: (gx, gy) Sobel della scena precomputati (evita un
+        Sobel full-frame per ogni match). Ritorna (cx, cy, angle_deg).
         """
-        if self.template_gray is None:
-            return cx, cy
-        h, w = self.template_gray.shape
-        scale = variant.scale
-        sw = max(16, int(round(w * scale)))
-        sh = max(16, int(round(h * scale)))
-        gray_s = cv2.resize(self.template_gray, (sw, sh), interpolation=cv2.INTER_LINEAR)
-        mask_src = (
-            self._train_mask if self._train_mask is not None
-            else np.full_like(self.template_gray, 255)
+        t, train_mask = self._get_view_template(getattr(variant, "view_idx", 0))
+        if t is None:
+            return cx, cy, angle_deg
+        mask_src = train_mask if train_mask is not None else np.full_like(t, 255)
+        gray_p, mask_p, center, diag = self._prepare_padded_template(
+            t, mask_src, variant.scale,
         )
-        mask_s = cv2.resize(mask_src, (sw, sh), interpolation=cv2.INTER_NEAREST)
-        diag = int(np.ceil(np.hypot(sh, sw))) + 6
-        py = (diag - sh) // 2; px = (diag - sw) // 2
-        gray_p = cv2.copyMakeBorder(
-            gray_s, py, diag - sh - py, px, diag - sw - px, cv2.BORDER_REPLICATE,
-        )
-        mask_p = cv2.copyMakeBorder(
-            mask_s, py, diag - sh - py, px, diag - sw - px,
-            cv2.BORDER_CONSTANT, value=0,
-        )
-        center = (diag / 2.0, diag / 2.0)
         M = cv2.getRotationMatrix2D(center, angle_deg, 1.0)
         gray_r = cv2.warpAffine(gray_p, M, (diag, diag),
                                  flags=cv2.INTER_LINEAR,
@@ -1241,51 +1185,98 @@ class LineShapeMatcher:
         mag_t = cv2.magnitude(gx_t, gy_t)
         _, bins_t = self._gradient(gray_r)
         fx, fy, _ = self._extract_features(mag_t, bins_t, mask_r)
-        if len(fx) < 4:
-            return cx, cy
-        n = len(fx)
-        ddx_t = (fx - center[0]).astype(np.float32)
-        ddy_t = (fy - center[1]).astype(np.float32)
-        gx_tf = np.array([gx_t[int(fy[i]), int(fx[i])] for i in range(n)], dtype=np.float32)
-        gy_tf = np.array([gy_t[int(fy[i]), int(fx[i])] for i in range(n)], dtype=np.float32)
-        mag_tf = np.hypot(gx_tf, gy_tf) + 1e-6
-        nx_t = gx_tf / mag_tf
-        ny_t = gy_tf / mag_tf
+        if len(fx) < 8:
+            return cx, cy, angle_deg
+        rx = (fx - center[0]).astype(np.float64)
+        ry = (fy - center[1]).astype(np.float64)
+        gxf = gx_t[fy, fx].astype(np.float64)
+        gyf = gy_t[fy, fx].astype(np.float64)
+        nm = np.hypot(gxf, gyf) + 1e-9
+        nx = gxf / nm
+        ny = gyf / nm
+
+        if scene_grad is not None:
+            gx_s, gy_s = scene_grad
+        else:
             gx_s = cv2.Sobel(scene_gray, cv2.CV_32F, 1, 0, ksize=3)
             gy_s = cv2.Sobel(scene_gray, cv2.CV_32F, 0, 1, ksize=3)
         H, W = scene_gray.shape
-        cur_cx, cur_cy = float(cx), float(cy)
-        for _ in range(n_iters):
-            xs = cur_cx + ddx_t
-            ys = cur_cy + ddy_t
-            xs_c = np.clip(xs, 0, W - 1.001)
-            ys_c = np.clip(ys, 0, H - 1.001)
+
+        def _bilin(g: np.ndarray, xs: np.ndarray, ys: np.ndarray) -> np.ndarray:
+            xs_c = np.clip(xs, 0.0, W - 1.001)
+            ys_c = np.clip(ys, 0.0, H - 1.001)
             x0 = xs_c.astype(np.int32); y0 = ys_c.astype(np.int32)
             ax = xs_c - x0; ay = ys_c - y0
-            def _bilin(g):
-                v00 = g[y0, x0]; v10 = g[y0, x0 + 1]
-                v01 = g[y0 + 1, x0]; v11 = g[y0 + 1, x0 + 1]
-                return ((1 - ax) * (1 - ay) * v00
-                        + ax * (1 - ay) * v10
-                        + (1 - ax) * ay * v01
-                        + ax * ay * v11)
-            sx_v = _bilin(gx_s)
-            sy_v = _bilin(gy_s)
-            mag_s = np.hypot(sx_v, sy_v) + 1e-6
-            nx_s = sx_v / mag_s
-            ny_s = sy_v / mag_s
-            w = np.minimum(mag_s, 255.0).astype(np.float32)
-            err_x = (nx_s - nx_t) * w
-            err_y = (ny_s - ny_t) * w
-            step_x = -float(err_x.sum()) / (w.sum() + 1e-6)
-            step_y = -float(err_y.sum()) / (w.sum() + 1e-6)
-            step_x = max(-1.0, min(1.0, step_x))
-            step_y = max(-1.0, min(1.0, step_y))
-            cur_cx += step_x
-            cur_cy += step_y
-            if abs(step_x) < 0.02 and abs(step_y) < 0.02:
+            return ((1 - ax) * (1 - ay) * g[y0, x0]
+                    + ax * (1 - ay) * g[y0, x0 + 1]
+                    + (1 - ax) * ay * g[y0 + 1, x0]
+                    + ax * ay * g[y0 + 1, x0 + 1])
+
+        t_offsets = np.array([-2.0, -1.0, 0.0, 1.0, 2.0])
+        n_feat = len(rx)
+        idx = np.arange(n_feat)
+        cur_cx, cur_cy, cur_ang = float(cx), float(cy), float(angle_deg)
+        for _ in range(n_iters):
+            px = cur_cx + rx
+            py = cur_cy + ry
+            # |grad| scena campionato a 5 offset lungo la normale di ogni
+            # feature; il picco sub-pixel lungo la normale e' la distanza
+            # firmata t_i dall'edge scena piu' vicino.
+            mags = np.empty((5, n_feat))
+            sxs = np.empty((5, n_feat))
+            sys_ = np.empty((5, n_feat))
+            for k, t_off in enumerate(t_offsets):
+                sx_v = _bilin(gx_s, px + t_off * nx, py + t_off * ny)
+                sy_v = _bilin(gy_s, px + t_off * nx, py + t_off * ny)
+                sxs[k] = sx_v; sys_[k] = sy_v
+                mags[k] = np.hypot(sx_v, sy_v)
+            k_best = np.argmax(mags, axis=0)
+            m_pk = mags[k_best, idx]
+            t_i = t_offsets[k_best]
+            # Fit parabolico sui picchi interni (k in 1..3)
+            interior = (k_best >= 1) & (k_best <= 3)
+            if interior.any():
+                ki = k_best[interior]; ii = idx[interior]
+                m_m = mags[ki - 1, ii]; m_0 = mags[ki, ii]; m_p = mags[ki + 1, ii]
+                denom = (m_m - 2.0 * m_0 + m_p)
+                off = np.where(np.abs(denom) > 1e-9,
+                               0.5 * (m_m - m_p) / (denom - 1e-12), 0.0)
+                t_i = t_i.astype(np.float64)
+                t_i[interior] += np.clip(off, -0.5, 0.5)
+            # Peso: |grad| al picco * allineamento direzione (mod π se no
+            # polarity). Feature senza edge (sotto weak_grad) escluse;
+            # picco sul bordo finestra = snap inaffidabile → dimezzato.
+            sx_pk = sxs[k_best, idx]; sy_pk = sys_[k_best, idx]
+            cos_al = (nx * sx_pk + ny * sy_pk) / (m_pk + 1e-9)
+            align = np.maximum(0.0, cos_al) if self.use_polarity else np.abs(cos_al)
+            wgt = np.minimum(m_pk, 255.0) * align * align
+            wgt[m_pk < self.weak_grad] = 0.0
+            wgt[~interior] *= 0.5
+            if float(wgt.sum()) < 1e-6:
                 break
-        return cur_cx, cur_cy
+            # LSQ pesato 3x3: A_i = [n_x, n_y, n_x·r_y - n_y·r_x]
+            a3 = nx * ry - ny * rx
+            A = np.stack([nx, ny, a3], axis=1)
+            Aw = A * wgt[:, None]
+            AtA = Aw.T @ A
+            Atb = Aw.T @ t_i.astype(np.float64)
+            try:
+                sol = np.linalg.solve(AtA + 1e-6 * np.eye(3), Atb)
+            except np.linalg.LinAlgError:
+                break
+            ddx = float(np.clip(sol[0], -1.5, 1.5))
+            ddy = float(np.clip(sol[1], -1.5, 1.5))
+            dth = float(np.clip(sol[2], -math.radians(1.5), math.radians(1.5)))
+            cur_cx += ddx
+            cur_cy += ddy
+            cur_ang += math.degrees(dth)
+            # Ruota offset e normali di dθ (convenzione R = [[c,s],[-s,c]])
+            c = math.cos(dth); s = math.sin(dth)
+            rx, ry = c * rx + s * ry, -s * rx + c * ry
+            nx, ny = c * nx + s * ny, -s * nx + c * ny
+            if abs(ddx) < 0.01 and abs(ddy) < 0.01 and abs(dth) < 1.7e-4:
+                break
+        return cur_cx, cur_cy, cur_ang
 
     def _verify_ncc(
         self, scene_gray: np.ndarray, cx: float, cy: float,
@@ -1370,15 +1361,21 @@ class LineShapeMatcher:
         coarse_stride: int = 1,
         scale_penalty: float = 0.0,
         search_roi: tuple[int, int, int, int] | None = None,
-        pyramid_propagate: bool = False,  # off di default: meno duplicati
-        propagate_topk: int = 4,
-        refine_pose_joint: bool = False,
+        # ON di default: full-res valutato solo in finestre locali attorno
+        # ai picchi top-level (costo ∝ candidati, non varianti × W × H).
+        # I duplicati che avevano fatto disattivare questa modalita' sono
+        # gestiti dalla NMS IoU poligonale post-refine.
+        pyramid_propagate: bool = True,
+        propagate_topk: int = 8,
+        refine_pose_joint: bool = False,  # deprecato: alias di refine_angle
         greediness: float = 0.0,
         batch_top: bool = False,
         nms_iou_threshold: float = 0.3,
         min_recall: float = 0.0,
         use_soft_score: bool = False,
-        subpixel_lm: bool = False,
+        # ON di default: least-squares finale (posizione + angolo) sui
+        # gradienti scena, precisione attesa <0.1 px / <0.1°.
+        subpixel_lm: bool = True,
         debug: bool = False,
         profile: bool = False,
     ) -> list[Match]:
@@ -1462,7 +1459,7 @@ class LineShapeMatcher:
         cached = self._scene_cache_get(cache_key) if cache_key else None
         if cached is not None:
             grays, spread_top, bit_active_top, density_top, spread0, \
-                bit_active_full, density_full, top = cached
+                bit_active_full, density_full, top, spread_fine = cached
         else:
             grays = [gray0]
             for _ in range(self.pyramid_levels - 1):
@@ -1478,22 +1475,47 @@ class LineShapeMatcher:
             spread0 = None
             bit_active_full = None
             density_full = None
+            spread_fine = None
         _checkpoint("spread_top")
         if nms_radius is None:
             nms_radius = max(8, min(self.template_size) // 2)
         # Pruning adattivo allo step angolare: con step piccolo (<= 3 deg)
-        # ci sono molte varianti vicine, gli score top-level sono ravvicinati
-        # e top_thresh*0.5 e' troppo aggressivo: scarta varianti valide che
-        # sarebbero state riprese al full-res. Stessa cosa per
-        # coarse_angle_factor (skip 1 ogni 2): con step fine non e' utile.
-        # Risultato osservato: precisione "veloce" 10° dava risultati
-        # migliori di "preciso" 2° proprio perche evitava il pruning.
+        # ci sono molte varianti vicine e gli score top-level sono
+        # ravvicinati: top_thresh*0.5 e' troppo aggressivo, scarta varianti
+        # valide che sarebbero state riprese al full-res.
+        # Il path windowed (pyramid_propagate) assume che il picco
+        # top-level localizzi la posizione entro il margine finestra.
+        # Su template ALLUNGATI (es. lama 40x280, o ROI parziale lungo un
+        # asse) lo score top-level ha un plateau lungo l'asse e il picco
+        # puo' essere lontano decine di px dal centro vero → le finestre
+        # tagliano fuori la posa giusta e il match muore in verify NCC.
+        # In quel caso si usa il full-scan esatto (costo maggiore ma
+        # nessuna perdita di recall).
+        if pyramid_propagate and self.template_size != (0, 0):
+            tw_t, th_t = self.template_size
+            if max(tw_t, th_t) / max(1, min(tw_t, th_t)) > 2.0:
+                pyramid_propagate = False
         eff_step = self._effective_angle_step()
         top_factor = self.top_score_factor
-        cf_eff = max(1, coarse_angle_factor)
         if eff_step <= 3.0:
             top_factor = max(top_factor, 0.7)
-            cf_eff = 1
+        # Coarse step angolare AUTO al top-level (Halcon-style): al livello
+        # top le feature distano R/2^top dal centro, quindi lo spread
+        # (raggio in px, costante per livello) tollera una rotazione
+        # ~atan(spread / (max_side_top/2)) — molto piu' ampia dello step
+        # richiesto a full-res. Si valuta al top 1 variante ogni cf_eff;
+        # le intermedie vengono riprese dall'espansione ai vicini.
+        # Es: template 160 px, 3 livelli, step 2° → tolleranza top ~11°
+        # → cf 6 → top-pruning ~6x piu' veloce a parita' di recall.
+        if self.template_size != (0, 0):
+            max_side_top = max(self.template_size) / (2 ** top)
+        else:
+            max_side_top = 64.0
+        step_top_tol = math.degrees(
+            math.atan2(float(self.spread_radius), max(8.0, max_side_top / 2.0))
+        )
+        cf_auto = int(np.clip(round(step_top_tol / max(eff_step, 1e-6)), 1, 8))
+        cf_eff = max(1, coarse_angle_factor, cf_auto)
         top_thresh = min_score * top_factor
         diag["top_thresh_used"] = float(top_thresh)
 
@@ -1549,7 +1571,6 @@ class LineShapeMatcher:
             dtype=bool,
         )
         if scene_bins.any():
-            n_scene_active = int(scene_bins.sum())
             # Soglia: variante deve avere >= 50% delle sue feature in bin
             # presenti nella scena. Sotto = score certamente < 0.5.
             pruned_idx_list = []
@@ -1598,16 +1619,25 @@ class LineShapeMatcher:
                 return vi, -1.0
             best = float(score.max())
             if pyramid_propagate and best > 0:
-                flat = score.ravel()
-                k = min(propagate_topk, flat.size)
-                idx = np.argpartition(-flat, k - 1)[:k]
+                # Picchi = MASSIMI LOCALI sopra soglia, non top-K pixel:
+                # su template allungati lo score top-level ha plateau
+                # estesi e i top-K pixel si concentrano tutti sulle 2-3
+                # istanze piu' forti, perdendo per sempre le altre.
+                # Soglia permissiva (0.5x): un picco scartato qui =
+                # istanza persa, un picco in piu' = solo una finestra
+                # extra di costo marginale (dedup via mark).
+                thr = top_thresh * 0.5
+                dil = cv2.dilate(score, np.ones((5, 5), np.uint8))
+                ys_l, xs_l = np.nonzero((score >= dil) & (score >= thr))
                 peaks: list[tuple[int, int, float]] = []
-                for i in idx:
-                    s = float(flat[i])
-                    if s < top_thresh * 0.7:
-                        continue
-                    yt, xt = int(i // score.shape[1]), int(i % score.shape[1])
-                    peaks.append((xt, yt, s))
+                if len(ys_l):
+                    vals = score[ys_l, xs_l]
+                    k = min(max(propagate_topk, 2 * max_matches), len(vals))
+                    sel = np.argpartition(-vals, k - 1)[:k]
+                    peaks = [
+                        (int(xs_l[i]), int(ys_l[i]), float(vals[i]))
+                        for i in sel
+                    ]
                 peaks_by_vi[vi] = peaks
             return vi, best
 
@@ -1664,6 +1694,13 @@ class LineShapeMatcher:
                 expanded.add(vi_n)
                 # Usa lo score del coarse come stima per il sort successivo
                 score_by_vi[vi_n] = max(score_by_vi.get(vi_n, 0.0), s_top)
+                # Propaga i picchi top-level del coarse anche ai vicini:
+                # l'oggetto e' nella stessa posizione (angolo ±step), quindi
+                # anche i vicini possono usare il path windowed invece del
+                # full-scan dell'intera scena (che dominava il costo full-res).
+                if (pyramid_propagate and vi_n != vi_c
+                        and peaks_by_vi.get(vi_c)):
+                    peaks_by_vi.setdefault(vi_n, []).extend(peaks_by_vi[vi_c])
         kept_variants: list[tuple[int, float]] = [
             (vi, score_by_vi[vi]) for vi in expanded
         ]
@@ -1690,54 +1727,63 @@ class LineShapeMatcher:
                     if (spread0 & (spread0.dtype.type(1) << b)).any())
             )
             density_full = _jit_popcount(spread0)
+            # Bitmap fine (raggio 1) per il refine: sulla bitmap dilatata
+            # lo score satura e il refine angolare/posizionale non vede
+            # alcun gradiente (vedi _refine_angle).
+            spread_fine = self._spread_bitmap(gray0, radius=1)
             # Salva cache scena complete
             if cache_key is not None:
                 self._scene_cache_put(cache_key, (
                     grays, spread_top, bit_active_top, density_top,
-                    spread0, bit_active_full, density_full, top,
+                    spread0, bit_active_full, density_full, top, spread_fine,
                 ))
         for sc in unique_scales:
             bg_cache_full[sc] = _bg_for_scale(density_full, sc, 1)
 
         # Margine in full-res attorno ad ogni peak top: copre incertezza
-        # downsampling (sf_top px) + spread_radius + slack per NMS.
-        propagate_margin = sf_top + self.spread_radius + max(8, nms_radius // 2)
+        # downsampling (2·sf_top px) + plateau radius del subpixel (10) +
+        # slack. NON serve includere nms_radius: la NMS lavora sui candidati
+        # estratti, non richiede score validi oltre il plateau del picco.
+        propagate_margin = 2 * sf_top + max(10, self.spread_radius) + 6
         H_full, W_full = spread0.shape
 
         def _full_score(vi: int) -> tuple[int, np.ndarray]:
             var = self.variants[vi]
             lvl0 = var.levels[0]
-            if not pyramid_propagate or vi not in peaks_by_vi or not peaks_by_vi[vi]:
+            peaks = peaks_by_vi.get(vi) if pyramid_propagate else None
+            margin = propagate_margin
+            if not peaks:
                 # Path legacy: scansiona intera scena
                 return vi, _jit_score_bitmap_rescored(
                     spread0, lvl0.dx, lvl0.dy, lvl0.bin, bit_active_full,
                     bg_cache_full[var.scale],
                 )
-            # Path piramide propagata: valuta solo crop locali attorno
-            # alle posizioni dei picchi top-level (riproiettati a full-res).
+            # Path piramide propagata: valuta solo finestre locali attorno
+            # ai picchi top-level (riproiettati a full-res). Il kernel
+            # windowed campiona lo spread dell'INTERA scena: chiamare il
+            # kernel su un crop trattava le feature fuori-crop come miss
+            # (template raggio > finestra → score ~0 ovunque, 0 match).
             score_full = np.zeros((H_full, W_full), dtype=np.float32)
             mark = np.zeros((H_full, W_full), dtype=bool)
             bg = bg_cache_full[var.scale]
-            for xt, yt, _s in peaks_by_vi[vi]:
+            for xt, yt, _s in peaks:
                 cx0 = xt * sf_top
                 cy0 = yt * sf_top
-                x_lo = max(0, cx0 - propagate_margin)
-                x_hi = min(W_full, cx0 + propagate_margin + 1)
-                y_lo = max(0, cy0 - propagate_margin)
-                y_hi = min(H_full, cy0 + propagate_margin + 1)
+                x_lo = max(0, cx0 - margin)
+                x_hi = min(W_full, cx0 + margin + 1)
+                y_lo = max(0, cy0 - margin)
+                y_hi = min(H_full, cy0 + margin + 1)
                 if x_hi <= x_lo or y_hi <= y_lo:
                     continue
                 if mark[y_lo:y_hi, x_lo:x_hi].all():
                     continue
-                # Crop spread + bg, valuta kernel sul crop
-                spread_crop = np.ascontiguousarray(spread0[y_lo:y_hi, x_lo:x_hi])
-                bg_crop = np.ascontiguousarray(bg[y_lo:y_hi, x_lo:x_hi])
-                score_crop = _jit_score_bitmap_rescored(
-                    spread_crop, lvl0.dx, lvl0.dy, lvl0.bin,
-                    bit_active_full, bg_crop,
+                score_win = _jit_score_bitmap_rescored_window(
+                    spread0, lvl0.dx, lvl0.dy, lvl0.bin,
+                    bit_active_full, bg,
+                    y_lo, x_lo, y_hi - y_lo, x_hi - x_lo,
                 )
                 score_full[y_lo:y_hi, x_lo:x_hi] = np.maximum(
-                    score_full[y_lo:y_hi, x_lo:x_hi], score_crop,
+                    score_full[y_lo:y_hi, x_lo:x_hi], score_win,
                 )
                 mark[y_lo:y_hi, x_lo:x_hi] = True
             return vi, score_full
@@ -1811,6 +1857,14 @@ class LineShapeMatcher:
         # Subpixel + refine + verify solo sui candidati pre-NMS (max pre_cap)
         kept: list[Match] = []
         tw, th = self.template_size
+        # Sobel scena precomputato una volta per il refine LM (prima era
+        # un Sobel full-frame per OGNI match).
+        scene_grad = None
+        if subpixel_lm and self.template_gray is not None and preliminary_int:
+            scene_grad = (
+                cv2.Sobel(gray0, cv2.CV_32F, 1, 0, ksize=3),
+                cv2.Sobel(gray0, cv2.CV_32F, 0, 1, ksize=3),
+            )
         for score, xi, yi, vi in preliminary_int:
             if subpixel and vi in score_maps:
                 cx_f, cy_f = self._subpixel_peak(
@@ -1821,12 +1875,10 @@ class LineShapeMatcher:
             var = self.variants[vi]
             ang_f = var.angle_deg
             score_f = score
-            if refine_pose_joint and self.template_gray is not None:
-                ang_f, score_f, cx_f, cy_f = self._refine_pose_joint(
-                    spread0, self.template_gray, cx_f, cy_f,
-                    var.angle_deg, var.scale, mask_full,
-                )
-            elif refine_angle and self.template_gray is not None:
+            # refine_pose_joint (Nelder-Mead) rimosso: valutava lo score a
+            # posizioni intere su bitmap satura (funzione a gradini piatta,
+            # il simplex terminava subito). Ora e' alias del refine standard.
+            if (refine_angle or refine_pose_joint) and self.template_gray is not None:
                 ang_f, score_f, cx_f, cy_f = self._refine_angle(
                     spread0, bit_active_full, self.template_gray, cx_f, cy_f,
                     var.angle_deg, var.scale, mask_full,
@@ -1835,14 +1887,15 @@ class LineShapeMatcher:
                     # del bin angolare della variante grezza.
                     search_radius=self._effective_angle_step(),
                     original_score=score,
+                    spread_fine=spread_fine,
                 )
-            # Halcon SubPixel='least_squares_high': refinement iterativo
-            # gradient-field per precisione 0.05 px (vs 0.5 quadratic 2D).
+            # Halcon SubPixel='least_squares_high': least-squares finale
+            # (posizione + angolo) sui gradienti scena, <0.1 px / <0.1°.
             if subpixel_lm and self.template_gray is not None:
-                cx_lm, cy_lm = self._subpixel_refine_lm(
-                    gray0, var, cx_f, cy_f, ang_f,
+                cx_lm, cy_lm, ang_lm = self._subpixel_refine_lm(
+                    gray0, var, cx_f, cy_f, ang_f, scene_grad=scene_grad,
                 )
-                cx_f, cy_f = float(cx_lm), float(cy_lm)
+                cx_f, cy_f, ang_f = float(cx_lm), float(cy_lm), float(ang_lm)
             # NCC verify (Halcon-style): se ncc_skip_above < 1.0 salta
             # il calcolo per shape-score gia alti. Default 1.01 = NCC sempre,
             # piu sicuro contro falsi positivi (lo shape-score satura facile).

pm2d/matcher.py

@@ -91,8 +91,16 @@ class EdgeShapeMatcher:
         a0, a1 = self.angle_range_deg
         if self.angle_step_deg <= 0 or a0 >= a1:
             return [float(a0)]
-        n = int(np.floor((a1 - a0) / self.angle_step_deg))
-        return [float(a0 + i * self.angle_step_deg) for i in range(n)]
+        # n+1 valori per includere l'estremo superiore del range: con il
+        # solo floor un range [0, 90] step 5 si fermava a 85° (off-by-one).
+        n = int(np.floor((a1 - a0) / self.angle_step_deg)) + 1
+        angles = [float(a0 + i * self.angle_step_deg) for i in range(n)]
+        if a1 - a0 >= 360.0:
+            # Range che copre il giro completo: a0+360° è la stessa pose di
+            # a0, escludi il duplicato (variante inutile in train/find).
+            eps = 1e-6
+            angles = [a for a in angles if a < a0 + 360.0 - eps]
+        return angles
 
     def train(self, template_bgr: np.ndarray) -> int:
         """Genera varianti per tutte le combinazioni (angolo, scala)."""
@@ -222,6 +230,14 @@ class EdgeShapeMatcher:
             for y, x in zip(ys, xs):
                 candidates.append((float(res[y, x]), int(x), int(y), ti))
 
+        # Cap candidati top-level: senza limite np.where con soglia bassa
+        # può generare migliaia di candidati, ognuno con un matchTemplate
+        # full-res nel refinement. Tieni solo i migliori per score.
+        max_candidates = max(1, max_matches * 10)
+        if len(candidates) > max_candidates:
+            candidates.sort(key=lambda c: -c[0])
+            candidates = candidates[:max_candidates]
+
         # Refinement a risoluzione piena: per ogni candidato top, finestra locale
         refined: list[tuple[float, int, int, int]] = []
         margin = sf + 4
@@ -294,6 +310,10 @@ class EdgeShapeMatcher:
         )
         arrays = {f"edge_{i}": t.edge for i, t in enumerate(self.templates)}
         arrays.update({f"mask_{i}": t.mask for i, t in enumerate(self.templates)})
+        # Persisti anche il grayscale originale: senza, l'overlay edge
+        # spariva dopo load() (template_gray restava None).
+        if self.template_gray is not None:
+            arrays["template_gray"] = self.template_gray
         np.savez_compressed(path, params=params, meta=meta, **arrays)
 
     @classmethod
@@ -312,6 +332,10 @@ class EdgeShapeMatcher:
             top_score_factor=float(p[12]) if len(p) > 12 else 0.6,
         )
         m.template_size = (int(p[8]), int(p[9]))
+        # Retrocompatibilità: modelli salvati prima non hanno template_gray
+        # (resta None: overlay edge non disponibile ma find() funziona).
+        if "template_gray" in z.files:
+            m.template_gray = z["template_gray"]
         meta = z["meta"]
         for i in range(len(meta)):
             m.templates.append(

pm2d/web/server.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 import hashlib
 import os
 import tempfile
+import threading
 import time
 import uuid
 from collections import OrderedDict
@@ -54,6 +55,7 @@ RECIPES_DIR.mkdir(exist_ok=True)
 
 from pm2d.line_matcher import LineShapeMatcher, Match
 from pm2d.auto_tune import auto_tune
+from pm2d.dxf import dxf_to_image
 
 
 WEB_DIR = Path(__file__).parent
@@ -64,14 +66,23 @@ STATIC_DIR.mkdir(exist_ok=True)
 CACHE_DIR = Path(tempfile.gettempdir()) / "pm2d_cache"
 CACHE_DIR.mkdir(exist_ok=True)
 
-# Cache in-memory (soft, ricaricata da disco se mancante)
-_IMG_CACHE: dict[str, np.ndarray] = {}
+# Cache in-memory (soft, ricaricata da disco se mancante).
+# LRU con capacità limitata: senza eviction le immagini si accumulavano
+# senza limite (leak di memoria su server long-running).
+_IMG_CACHE: OrderedDict[str, np.ndarray] = OrderedDict()
+_IMG_CACHE_SIZE = 64
 
 # Cache matcher addestrati: (roi_hash, params_hash) -> LineShapeMatcher
 # LRU con capacità limitata
 _MATCHER_CACHE: OrderedDict = OrderedDict()
 _MATCHER_CACHE_SIZE = 8
 
+# Lock globale matcher: gli endpoint girano nel threadpool FastAPI ma i
+# matcher condivisi (_MATCHER_CACHE, _RECIPE_MATCHERS) mutano stato interno
+# durante train()/find(). Serializzare il matching è la soluzione semplice
+# e corretta (un lock per-ricetta sarebbe over-engineering).
+_MATCHER_LOCK = threading.Lock()
+
 
 def _matcher_cache_key(roi: np.ndarray, tech: dict) -> str:
     h = hashlib.md5()
@@ -81,7 +92,10 @@ def _matcher_cache_key(roi: np.ndarray, tech: dict) -> str:
                 "min_feature_spacing",
                 "angle_min", "angle_max", "angle_step",
                 "scale_min", "scale_max", "scale_step",
-                "spread_radius", "pyramid_levels")
+                "spread_radius", "pyramid_levels",
+                # ROI poligonale: la mask cambia il training a parità di
+                # bbox → deve invalidare la cache (None = ROI rettangolare)
+                "roi_poly")
     for k in relevant:
         h.update(f"{k}={tech.get(k)}".encode())
     h.update(f"shape={roi.shape}".encode())
@@ -102,23 +116,32 @@ def _cache_put_matcher(key: str, matcher) -> None:
         _MATCHER_CACHE.popitem(last=False)
 
 
+def _img_cache_put(key: str, value: np.ndarray) -> None:
+    """Inserisce in _IMG_CACHE con eviction LRU (cap _IMG_CACHE_SIZE)."""
+    _IMG_CACHE[key] = value
+    _IMG_CACHE.move_to_end(key)
+    while len(_IMG_CACHE) > _IMG_CACHE_SIZE:
+        _IMG_CACHE.popitem(last=False)
+
+
 def _store_image(img: np.ndarray) -> str:
     iid = uuid.uuid4().hex[:12]
     cv2.imwrite(str(CACHE_DIR / f"{iid}.png"), img)
-    _IMG_CACHE[iid] = img
+    _img_cache_put(iid, img)
     return iid
 
 
 def _load_image(iid: str) -> np.ndarray | None:
     cached = _IMG_CACHE.get(iid)
     if cached is not None:
+        _IMG_CACHE.move_to_end(iid)  # LRU touch
         return cached
     p = CACHE_DIR / f"{iid}.png"
     if not p.exists():
         return None
     img = cv2.imread(str(p))
     if img is not None:
-        _IMG_CACHE[iid] = img
+        _img_cache_put(iid, img)
     return img
 
 app = FastAPI(title="PM2D Webapp", version="1.0.0")
@@ -131,6 +154,72 @@ def _encode_png(img: np.ndarray) -> bytes:
     return buf.tobytes()
 
 
+def _clamp_roi(x: int, y: int, w: int, h: int,
+               img_w: int, img_h: int) -> tuple[int, int, int, int]:
+    """Clampa la ROI dentro i limiti immagine.
+
+    Una ROI fuori immagine causava slice vuote → crash 500 negli endpoint
+    che non clampavano. Solleva 400 se la ROI risultante è degenere
+    (lato < 16 px: sotto questa soglia il train non estrae abbastanza
+    edge feature e produce 0 varianti → find() esplode con 500).
+    """
+    x = max(0, min(int(x), img_w - 1))
+    y = max(0, min(int(y), img_h - 1))
+    w = min(int(w), img_w - x)
+    h = min(int(h), img_h - y)
+    if w < 16 or h < 16:
+        raise HTTPException(
+            400, f"ROI fuori immagine o degenere: [{x}, {y}, {w}, {h}] "
+                 f"su immagine {img_w}x{img_h} (lato minimo 16 px)")
+    return x, y, w, h
+
+
+def _poly_bbox_mask(
+    roi_poly: list[list[float]], img_w: int, img_h: int,
+) -> tuple[int, int, int, int, np.ndarray]:
+    """Valida roi_poly (vertici [x, y] in coordinate IMMAGINE) e ritorna
+    (x, y, w, h, mask): bbox del poligono clampato con _clamp_roi e mask
+    uint8 (255 dentro il poligono) nel sistema di coordinate della ROI.
+
+    Solleva 400 se il poligono ha <3 punti o area degenere.
+    """
+    pts = np.asarray(roi_poly, dtype=np.float64)
+    if pts.ndim != 2 or pts.shape[1] != 2 or pts.shape[0] < 3:
+        raise HTTPException(
+            400, "roi_poly non valido: servono almeno 3 vertici [x, y]")
+    # Area con formula shoelace: poligoni collineari/degeneri → 400
+    px_, py_ = pts[:, 0], pts[:, 1]
+    area = 0.5 * abs(np.dot(px_, np.roll(py_, 1)) - np.dot(py_, np.roll(px_, 1)))
+    if area < 16.0:
+        raise HTTPException(
+            400, f"roi_poly degenere: area {area:.1f} px² troppo piccola")
+    x0 = int(np.floor(px_.min())); y0 = int(np.floor(py_.min()))
+    bw = int(np.ceil(px_.max())) - x0
+    bh = int(np.ceil(py_.max())) - y0
+    x, y, w, h = _clamp_roi(x0, y0, bw, bh, img_w, img_h)
+    # Mask nel sistema ROI: vertici ritraslati di (-x, -y)
+    mask = np.zeros((h, w), dtype=np.uint8)
+    local = np.round(pts - [x, y]).astype(np.int32)
+    cv2.fillPoly(mask, [local], 255)
+    if not mask.any():
+        raise HTTPException(
+            400, "roi_poly fuori immagine: nessun pixel utile nella mask")
+    return x, y, w, h, mask
+
+
+def _check_trained(m: "LineShapeMatcher", n_variants: int) -> None:
+    """Solleva 422 se il train non ha prodotto varianti.
+
+    Succede con ROI senza contrasto (sfondo uniforme) o troppo piccola:
+    senza questo check il find() successivo esplode con RuntimeError → 500.
+    """
+    if n_variants <= 0 or not m.variants:
+        raise HTTPException(
+            422, "La ROI non contiene abbastanza edge feature per il "
+                 "training (zona troppo uniforme o piccola): scegliere "
+                 "una regione con contorni netti")
+
+
 def _draw_matches(scene: np.ndarray, matches: list[Match],
                    template_gray: np.ndarray | None,
                    matcher: "LineShapeMatcher | None" = None) -> np.ndarray:
@@ -172,17 +261,29 @@ def _draw_matches(scene: np.ndarray, matches: list[Match],
             # `center = (diag / 2.0, diag / 2.0)` (no -1). Usare (tw-1)/2
             # introduceva uno shift di 0.5px per template di lato pari.
             cx_t = tw / 2.0; cy_t = th / 2.0
+            # Lavora su un CROP locale della scena di lato = diagonale del
+            # template ruotato+scalato (+margine), come _verify_ncc: warp
+            # + Sobel sull'INTERA scena per ogni match erano O(W·H) cadauno
+            # (costosissimo su scene grandi con molti match).
+            diag = int(np.ceil(np.hypot(tw, th) * m.scale)) + 8
+            x0 = int(round(m.cx)) - diag // 2
+            y0 = int(round(m.cy)) - diag // 2
+            gx0 = max(0, x0); gy0 = max(0, y0)
+            gx1 = min(W_scene, x0 + diag); gy1 = min(H_scene, y0 + diag)
+            cw, ch_ = gx1 - gx0, gy1 - gy0
+            if cw >= 3 and ch_ >= 3:
                 M = cv2.getRotationMatrix2D((cx_t, cy_t), m.angle_deg, m.scale)
-            M[0, 2] += m.cx - cx_t
-            M[1, 2] += m.cy - cy_t
+                # Porta il centro template a (m.cx - gx0, m.cy - gy0) del crop
+                M[0, 2] += (m.cx - gx0) - cx_t
+                M[1, 2] += (m.cy - gy0) - cy_t
                 warped_gray = cv2.warpAffine(
-                t, M, (W_scene, H_scene),
+                    t, M, (cw, ch_),
                     flags=cv2.INTER_LINEAR, borderValue=0)
                 # Maschera: train_mask se disponibile, altrimenti rettangolo pieno
                 mask_src = (matcher._train_mask if matcher._train_mask is not None
                             else np.full((th, tw), 255, dtype=np.uint8))
                 warped_mask = cv2.warpAffine(
-                mask_src, M, (W_scene, H_scene),
+                    mask_src, M, (cw, ch_),
                     flags=cv2.INTER_NEAREST, borderValue=0)
                 # Erode minimo (3x3) per togliere SOLO artefatti border-padding
                 # (~1px di bordo nero da warpAffine borderValue=0). Erode piu'
@@ -197,11 +298,16 @@ def _draw_matches(scene: np.ndarray, matches: list[Match],
                     edge_mask = mag >= matcher.strong_grad
                 edge_mask = edge_mask & (warped_mask > 0)
                 if edge_mask.any():
-                edge_overlay = np.zeros_like(out)
+                    # Edge ritraslati nel sistema scena: blend solo sul crop
+                    # (addWeighted lascia invariati i pixel con overlay nullo,
+                    # quindi l'output visivo è identico al full-frame).
+                    sub = out[gy0:gy1, gx0:gx1]
+                    edge_overlay = np.zeros_like(sub)
                     # Ciano (cambiato da verde): non collide col verde dell'asse
                     # Y dell'UCS che altrimenti scompariva nell'overlay edge.
                     edge_overlay[edge_mask] = (255, 200, 0)  # ciano (BGR)
-                out = cv2.addWeighted(out, 1.0, edge_overlay, 0.6, 0)
+                    out[gy0:gy1, gx0:gx1] = cv2.addWeighted(
+                        sub, 1.0, edge_overlay, 0.6, 0)
         L = max(20, int(L_base * m.scale))
         # X axis = rotazione di (1, 0) con cv2 matrix → (cos, -sin)
         x_end = (int(cx + L * ca), int(cy - L * sa))
@@ -234,6 +340,10 @@ class MatchParams(BaseModel):
     model_id: str
     scene_id: str
     roi: list[int]  # [x, y, w, h] nell'immagine modello
+    # ROI poligonale opzionale: vertici [x, y] in coordinate IMMAGINE
+    # (min 3 punti). Se presente, il bbox del poligono sostituisce `roi`
+    # e il training usa la mask del poligono.
+    roi_poly: list[list[float]] | None = None
     angle_min: float = 0.0
     angle_max: float = 360.0
     angle_step: float = 5.0
@@ -246,7 +356,9 @@ class MatchParams(BaseModel):
     num_features: int = 96
     weak_grad: float = 30.0
     strong_grad: float = 60.0
-    spread_radius: int = 5
+    # 4 allineato col default del matcher: raggio 5 peggiora la precisione
+    # di rotazione (spread troppo largo appiattisce il picco angolare).
+    spread_radius: int = 4
     pyramid_levels: int = 3
     verify_threshold: float = 0.4
 
@@ -313,6 +425,8 @@ class SimpleMatchParams(BaseModel):
     model_id: str
     scene_id: str
     roi: list[int]
+    # ROI poligonale opzionale (vedi MatchParams.roi_poly)
+    roi_poly: list[list[float]] | None = None
     tipo: str = "intero"              # "intero" | "parziale"
     simmetria: str = "nessuna"        # chiave SYMMETRY_TO_ANGLE_MAX
     scala: str = "fissa"              # chiave SCALE_PRESETS
@@ -407,7 +521,13 @@ def _simple_to_technical(
         "min_score": p.min_score,
         "max_matches": p.max_matches,
         "nms_radius": 0,
-        "verify_threshold": FILTRO_FP_MAP.get(p.filtro_fp, 0.35),
+        # Fallback = livello "medio" della mappa (no valore hardcoded
+        # che divergerebbe se la mappa cambia).
+        "verify_threshold": FILTRO_FP_MAP.get(p.filtro_fp, FILTRO_FP_MAP["medio"]),
+        # "off" deve disabilitare DAVVERO il verify NCC: passare solo
+        # verify_threshold=0.0 lascerebbe attivo il calcolo NCC (che può
+        # comunque scartare match con score negativo / patch uniformi).
+        "verify_ncc": p.filtro_fp != "off",
         "scale_penalty": p.penalita_scala,
     }
 
@@ -526,6 +646,26 @@ async def upload(file: UploadFile = File(...)):
     return UploadResp(id=iid, width=img.shape[1], height=img.shape[0])
 
 
+@app.post("/upload_dxf", response_model=UploadResp)
+async def upload_dxf(file: UploadFile = File(...), size: int = 512):
+    """Upload DXF: rasterizza il CAD in template grayscale e lo salva
+    nella cache immagini come un normale upload.
+
+    Query param `size` = lato del canvas (clamp 128..2048).
+    """
+    size = max(128, min(2048, int(size)))
+    data = await file.read()
+    try:
+        gray = dxf_to_image(data, target_size=size)
+    except ValueError as e:
+        raise HTTPException(400, f"DXF non valido: {e}")
+    # _store_image salva PNG e gli endpoint a valle (cvtColor BGR2GRAY su
+    # roi_img, _load_image con IMREAD_COLOR) si aspettano 3 canali → BGR.
+    img = cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR)
+    iid = _store_image(img)
+    return UploadResp(id=iid, width=img.shape[1], height=img.shape[0])
+
+
 @app.get("/image/{iid}/raw")
 def image_raw(iid: str):
     img = _load_image(iid)
@@ -540,10 +680,14 @@ def match(p: MatchParams):
     scene = _load_image(p.scene_id)
     if model is None or scene is None:
         raise HTTPException(404, "Immagini non trovate")
+    # ROI poligonale: bbox derivato dal poligono + mask per il training
+    train_mask = None
+    if p.roi_poly is not None:
+        x, y, w, h, train_mask = _poly_bbox_mask(
+            p.roi_poly, model.shape[1], model.shape[0])
+    else:
         x, y, w, h = p.roi
-    x = max(0, x); y = max(0, y)
-    w = max(1, min(w, model.shape[1] - x))
-    h = max(1, min(h, model.shape[0] - y))
+        x, y, w, h = _clamp_roi(x, y, w, h, model.shape[1], model.shape[0])
     roi_img = model[y:y + h, x:x + w]
 
     tech_for_cache = {
@@ -555,8 +699,13 @@ def match(p: MatchParams):
         "scale_step": p.scale_step,
         "spread_radius": p.spread_radius,
         "pyramid_levels": p.pyramid_levels,
+        # Tuple per repr stabile nella cache key (None = rettangolare)
+        "roi_poly": (tuple(map(tuple, p.roi_poly))
+                     if p.roi_poly is not None else None),
     }
     key = _matcher_cache_key(roi_img, tech_for_cache)
+    # Lock globale: matcher condivisi tra thread del pool FastAPI
+    with _MATCHER_LOCK:
         m = _cache_get_matcher(key)
         if m is None:
             m = LineShapeMatcher(
@@ -569,7 +718,8 @@ def match(p: MatchParams):
                 spread_radius=p.spread_radius,
                 pyramid_levels=p.pyramid_levels,
             )
-        t0 = time.time(); n = m.train(roi_img); t_train = time.time() - t0
+            t0 = time.time(); n = m.train(roi_img, train_mask); t_train = time.time() - t0
+            _check_trained(m, n)
             _cache_put_matcher(key, m)
         else:
             n = len(m.variants); t_train = 0.0
@@ -578,6 +728,8 @@ def match(p: MatchParams):
         matches = m.find(
             scene, min_score=p.min_score, max_matches=p.max_matches,
             nms_radius=nms, verify_threshold=p.verify_threshold,
+            # Soglia 0 = filtro FP disattivato: skippa proprio il calcolo NCC
+            verify_ncc=p.verify_threshold > 0.0,
         )
         t_find = time.time() - t0
 
@@ -609,18 +761,27 @@ def match_simple(p: SimpleMatchParams):
     scene = _load_image(p.scene_id)
     if model is None or scene is None:
         raise HTTPException(404, "Immagini non trovate")
+    # ROI poligonale: bbox derivato dal poligono + mask per il training
+    train_mask = None
+    if p.roi_poly is not None:
+        x, y, w, h, train_mask = _poly_bbox_mask(
+            p.roi_poly, model.shape[1], model.shape[0])
+    else:
         x, y, w, h = p.roi
-    x = max(0, x); y = max(0, y)
-    w = max(1, min(w, model.shape[1] - x))
-    h = max(1, min(h, model.shape[0] - y))
+        x, y, w, h = _clamp_roi(x, y, w, h, model.shape[1], model.shape[0])
     roi_img = model[y:y + h, x:x + w]
 
     tech = _simple_to_technical(p, roi_img)
+    # Tuple per repr stabile nella cache key (None = rettangolare)
+    tech["roi_poly"] = (tuple(map(tuple, p.roi_poly))
+                        if p.roi_poly is not None else None)
 
     key = _matcher_cache_key(roi_img, tech)
     # Halcon-mode init params: incidono sul training, includere in cache key
     halcon_init_key = f"|pol={p.use_polarity}|gpu={p.use_gpu}"
     key = key + halcon_init_key
+    # Lock globale: matcher condivisi tra thread del pool FastAPI
+    with _MATCHER_LOCK:
         m = _cache_get_matcher(key)
         if m is None:
             m = LineShapeMatcher(
@@ -636,7 +797,8 @@ def match_simple(p: SimpleMatchParams):
                 use_polarity=p.use_polarity,
                 use_gpu=p.use_gpu,
             )
-        t0 = time.time(); n = m.train(roi_img); t_train = time.time() - t0
+            t0 = time.time(); n = m.train(roi_img, train_mask); t_train = time.time() - t0
+            _check_trained(m, n)
             _cache_put_matcher(key, m)
         else:
             n = len(m.variants); t_train = 0.0
@@ -646,6 +808,8 @@ def match_simple(p: SimpleMatchParams):
         matches = m.find(
             scene, min_score=tech["min_score"], max_matches=tech["max_matches"],
             nms_radius=nms, verify_threshold=tech["verify_threshold"],
+            # filtro_fp="off" → verify NCC davvero disabilitato
+            verify_ncc=tech.get("verify_ncc", True),
             scale_penalty=tech.get("scale_penalty", 0.0),
             # Halcon-mode flags
             min_recall=p.min_recall,
@@ -681,6 +845,7 @@ def tune(p: TuneParams):
     if model is None:
         raise HTTPException(404, "Immagine non trovata")
     x, y, w, h = p.roi
+    x, y, w, h = _clamp_roi(x, y, w, h, model.shape[1], model.shape[0])
     roi_img = model[y:y + h, x:x + w]
     t = auto_tune(roi_img)
     # Esponi parametri tecnici + meta diagnostica (_self_score, _validation,
@@ -694,6 +859,8 @@ class SaveRecipeParams(BaseModel):
     model_id: str
     scene_id: str | None = None
     roi: list[int]
+    # ROI poligonale opzionale (vedi MatchParams.roi_poly)
+    roi_poly: list[list[float]] | None = None
     # Riusa stessi param simple per training equivalente
     tipo: str = "intero"
     simmetria: str = "nessuna"
@@ -813,7 +980,14 @@ def save_recipe(p: SaveRecipeParams):
     model = _load_image(p.model_id)
     if model is None:
         raise HTTPException(404, "Modello non trovato")
+    # ROI poligonale: bbox derivato dal poligono + mask per il training
+    train_mask = None
+    if p.roi_poly is not None:
+        x, y, w, h, train_mask = _poly_bbox_mask(
+            p.roi_poly, model.shape[1], model.shape[0])
+    else:
         x, y, w, h = p.roi
+        x, y, w, h = _clamp_roi(x, y, w, h, model.shape[1], model.shape[0])
     roi_img = model[y:y + h, x:x + w]
     sp = SimpleMatchParams(
         model_id=p.model_id, scene_id=p.scene_id or p.model_id, roi=p.roi,
@@ -838,7 +1012,10 @@ def save_recipe(p: SaveRecipeParams):
         use_polarity=p.use_polarity,
         use_gpu=p.use_gpu,
     )
-    m.train(roi_img)
+    # Lock globale: serializza il training pesante col matching in corso
+    with _MATCHER_LOCK:
+        n_var = m.train(roi_img, train_mask)
+    _check_trained(m, n_var)
     safe_name = "".join(c for c in p.name if c.isalnum() or c in "._-")
     if not safe_name:
         raise HTTPException(400, "Nome ricetta non valido")
@@ -864,6 +1041,14 @@ _RECIPE_MATCHERS: OrderedDict = OrderedDict()
 _RECIPE_MATCHERS_SIZE = 4
 
 
+def _recipe_matchers_put(name: str, matcher: LineShapeMatcher) -> None:
+    """Inserisce in _RECIPE_MATCHERS con eviction LRU (cap _RECIPE_MATCHERS_SIZE)."""
+    _RECIPE_MATCHERS[name] = matcher
+    _RECIPE_MATCHERS.move_to_end(name)
+    while len(_RECIPE_MATCHERS) > _RECIPE_MATCHERS_SIZE:
+        _RECIPE_MATCHERS.popitem(last=False)
+
+
 @app.post("/recipes/{name}/load")
 def load_recipe(name: str):
     """Carica ricetta .npz e popola cache matcher in memoria.
@@ -878,10 +1063,8 @@ def load_recipe(name: str):
     if not path.is_file():
         raise HTTPException(404, f"Ricetta non trovata: {safe_name}")
     m = LineShapeMatcher.load_model(str(path))
-    _RECIPE_MATCHERS[safe_name] = m
-    _RECIPE_MATCHERS.move_to_end(safe_name)
-    while len(_RECIPE_MATCHERS) > _RECIPE_MATCHERS_SIZE:
-        _RECIPE_MATCHERS.popitem(last=False)
+    with _MATCHER_LOCK:
+        _recipe_matchers_put(safe_name, m)
     return {
         "name": safe_name,
         "n_variants": len(m.variants),
@@ -905,7 +1088,9 @@ class RecipeMatchParams(BaseModel):
     greediness: float = 0.0
     refine_pose_joint: bool = False
     search_roi: list[int] | None = None
-    verify_threshold: float = 0.5
+    # Allineato a MatchParams.verify_threshold (0.4): valori divergenti
+    # davano risultati diversi tra /match e /match_recipe a parità di scena.
+    verify_threshold: float = 0.4
     scale_penalty: float = 0.0
 
 
@@ -913,23 +1098,30 @@ class RecipeMatchParams(BaseModel):
 def match_recipe(p: RecipeMatchParams):
     """Match con ricetta pre-trained: zero training, solo find."""
     safe_name = p.recipe if p.recipe.endswith(".npz") else f"{p.recipe}.npz"
-    m = _RECIPE_MATCHERS.get(safe_name)
-    if m is None:
-        # Auto-load on demand
-        path = RECIPES_DIR / safe_name
-        if not path.is_file():
-            raise HTTPException(404, f"Ricetta non trovata: {safe_name}")
-        m = LineShapeMatcher.load_model(str(path))
-        _RECIPE_MATCHERS[safe_name] = m
     scene = _load_image(p.scene_id)
     if scene is None:
         raise HTTPException(404, "Scena non trovata")
     search_roi_t = tuple(p.search_roi) if p.search_roi else None
+    # Lock globale: matcher condivisi tra thread del pool FastAPI
+    with _MATCHER_LOCK:
+        m = _RECIPE_MATCHERS.get(safe_name)
+        if m is not None:
+            _RECIPE_MATCHERS.move_to_end(safe_name)  # LRU touch
+        else:
+            # Auto-load on demand: stessa eviction LRU di load_recipe
+            # (senza cap la cache cresceva senza limite)
+            path = RECIPES_DIR / safe_name
+            if not path.is_file():
+                raise HTTPException(404, f"Ricetta non trovata: {safe_name}")
+            m = LineShapeMatcher.load_model(str(path))
+            _recipe_matchers_put(safe_name, m)
         t0 = time.time()
         matches = m.find(
             scene,
             min_score=p.min_score, max_matches=p.max_matches,
             verify_threshold=p.verify_threshold,
+            # Soglia 0 = filtro FP disattivato: skippa proprio il calcolo NCC
+            verify_ncc=p.verify_threshold > 0.0,
             scale_penalty=p.scale_penalty,
             min_recall=p.min_recall,
             use_soft_score=p.use_soft_score,

pm2d/web/static/app.js

@@ -20,6 +20,10 @@ const state = {
   model: null, scene: null, roi: null, drag: null,
   matches: [], annotatedImg: null,
   active_recipe: null,  // V: ricetta caricata (string nome) o null
+  // ROI poligonale: vertici [x, y] in coordinate immagine modello
+  polyMode: false, polyPts: [], polyClosed: false,
+  // Export JSON: ultimo match completo (params + risposta)
+  lastMatch: null,
 };
 
 // ---------- Forms ----------
@@ -148,6 +152,15 @@ async function uploadToFolder(file) {
   return await r.json();
 }
 
+async function uploadDxf(file) {
+  // DXF: rasterizzato server-side in template grayscale (vedi pm2d/dxf.py)
+  const fd = new FormData();
+  fd.append("file", file);
+  const r = await fetch("/upload_dxf", { method: "POST", body: fd });
+  if (!r.ok) throw new Error(await r.text());
+  return await r.json();
+}
+
 async function refreshPickers() {
   const {files, dir} = await fetchImagesList();
   buildThumbPicker("picker-model", files, onSelectModel);
@@ -222,6 +235,7 @@ async function onSelectModel(filename) {
     const img = await loadImage(`/image/${meta.id}/raw`);
     state.model = { id: meta.id, w: meta.width, h: meta.height, img };
     state.roi = null;
+    state.polyPts = []; state.polyClosed = false;  // B: scarta poligono stale
     document.getElementById("roi-info").textContent = "ROI: (nessuna)";
     setStatus(`Modello: ${filename} ${meta.width}x${meta.height} — trascina ROI`);
     renderModel();
@@ -262,12 +276,36 @@ function renderModel() {
   state.model.scale = fit.sc;
   state.model.ox = fit.ox; state.model.oy = fit.oy;
   ctx.drawImage(state.model.img, fit.ox, fit.oy, fit.dw, fit.dh);
-  if (state.roi) {
+  if (state.roi && !state.polyMode) {
     const [x, y, w, h] = state.roi;
     ctx.strokeStyle = "#00ff80"; ctx.lineWidth = 2;
     ctx.strokeRect(fit.ox + x * fit.sc, fit.oy + y * fit.sc,
                    w * fit.sc, h * fit.sc);
   }
+  // ROI poligonale: path aperto giallo, chiuso verde con fill semitrasparente
+  if (state.polyMode && state.polyPts.length > 0) {
+    ctx.beginPath();
+    state.polyPts.forEach(([px, py], i) => {
+      const cx = fit.ox + px * fit.sc;
+      const cy = fit.oy + py * fit.sc;
+      if (i === 0) ctx.moveTo(cx, cy); else ctx.lineTo(cx, cy);
+    });
+    if (state.polyClosed) {
+      ctx.closePath();
+      ctx.fillStyle = "rgba(0, 255, 128, 0.18)";
+      ctx.fill();
+      ctx.strokeStyle = "#00ff80";
+    } else {
+      ctx.strokeStyle = "#ffff00";
+    }
+    ctx.lineWidth = 2;
+    ctx.stroke();
+    // Vertici come quadratini
+    ctx.fillStyle = state.polyClosed ? "#00ff80" : "#ffff00";
+    for (const [px, py] of state.polyPts) {
+      ctx.fillRect(fit.ox + px * fit.sc - 2, fit.oy + py * fit.sc - 2, 4, 4);
+    }
+  }
   if (state.drag) {
     ctx.strokeStyle = "#ffff00";
     ctx.setLineDash([4, 2]); ctx.lineWidth = 2;
@@ -301,10 +339,35 @@ function setupROI() {
   const cnv = document.getElementById("c-model");
   cnv.addEventListener("mousedown", (e) => {
     if (!state.model) return;
+    if (state.polyMode) return;  // poly mode: gestito da click/dblclick
     const p = canvasPos(cnv, e);
     state.drag = { x0: p.x, y0: p.y, x1: p.x, y1: p.y };
     renderModel();
   });
+  // ROI poligonale: click aggiunge vertice, doppio click chiude
+  cnv.addEventListener("click", (e) => {
+    if (!state.model || !state.polyMode || state.polyClosed) return;
+    const m = state.model;
+    const p = canvasPos(cnv, e);
+    const ix = (p.x - m.ox) / m.scale;
+    const iy = (p.y - m.oy) / m.scale;
+    if (ix < 0 || iy < 0 || ix > m.w || iy > m.h) return;  // fuori immagine
+    const last = state.polyPts[state.polyPts.length - 1];
+    // Dedup: il dblclick genera anche 2 click ravvicinati
+    if (last && Math.hypot(ix - last[0], iy - last[1]) < 3) return;
+    state.polyPts.push([
+      Math.max(0, Math.min(Math.round(ix), m.w - 1)),
+      Math.max(0, Math.min(Math.round(iy), m.h - 1)),
+    ]);
+    document.getElementById("roi-info").textContent =
+      `Poligono: ${state.polyPts.length} vertici (doppio click o "Chiudi" per chiudere)`;
+    renderModel();
+  });
+  cnv.addEventListener("dblclick", (e) => {
+    if (!state.polyMode) return;
+    e.preventDefault();
+    closePoly();
+  });
   cnv.addEventListener("mousemove", (e) => {
     if (!state.drag) return;
     const p = canvasPos(cnv, e);
@@ -331,6 +394,41 @@ function setupROI() {
   });
 }
 
+// ---------- ROI poligonale ----------
+function closePoly() {
+  if (!state.polyMode || state.polyClosed) return;
+  if (state.polyPts.length < 3) {
+    setStatus("Servono almeno 3 vertici per chiudere il poligono");
+    return;
+  }
+  state.polyClosed = true;
+  // ROI = bounding box del poligono (il server riceve anche roi_poly)
+  const xs = state.polyPts.map((p) => p[0]);
+  const ys = state.polyPts.map((p) => p[1]);
+  const x0 = Math.min(...xs), y0 = Math.min(...ys);
+  const w = Math.max(...xs) - x0, h = Math.max(...ys) - y0;
+  state.roi = [x0, y0, Math.max(1, w), Math.max(1, h)];
+  document.getElementById("roi-info").textContent =
+    `Poligono: ${state.polyPts.length} vertici, bbox ${w}x${h} @ (${x0}, ${y0})`;
+  renderModel();
+}
+
+function resetPoly() {
+  state.polyPts = [];
+  state.polyClosed = false;
+  state.roi = null;
+  document.getElementById("roi-info").textContent = state.polyMode
+    ? "Poligono: clicca sul modello per aggiungere vertici"
+    : "ROI: (nessuna)";
+  renderModel();
+}
+
+function getRoiPoly() {
+  // Poligono valido solo se in modalità poly e chiuso
+  return (state.polyMode && state.polyClosed && state.polyPts.length >= 3)
+    ? state.polyPts : null;
+}
+
 // ---------- Match action ----------
 async function doMatchRecipe() {
   if (!state.scene) { setStatus("Carica scena"); return; }
@@ -352,6 +450,12 @@ async function doMatchRecipe() {
   if (!r.ok) { setStatus(`Errore: ${await r.text()}`); return; }
   const data = await r.json();
   state.matches = data.matches;
+  // C: salva tutto per "Esporta JSON"
+  state.lastMatch = {
+    endpoint: "/match_recipe", params: body, response: data,
+    image_id: state.scene.id,
+  };
+  document.getElementById("btn-export-json").disabled = false;
   state.annotatedImg = await loadImage(
     `/image/${data.annotated_id}/raw?t=${Date.now()}`);
   renderScene();
@@ -371,7 +475,11 @@ async function doMatch() {
   }
   if (!state.model) { setStatus("Carica modello"); return; }
   if (!state.scene) { setStatus("Carica scena"); return; }
+  if (state.polyMode && !state.polyClosed) {
+    setStatus("Chiudi il poligono (doppio click o bottone Chiudi)"); return;
+  }
   if (!state.roi) { setStatus("Seleziona ROI sul modello"); return; }
+  const roiPoly = getRoiPoly();
   const user = readUserParams();
   const adv = readAdvancedOverrides();
   setStatus("Match in corso...");
@@ -397,6 +505,7 @@ async function doMatch() {
     const angMax = SYM_MAP[user.simmetria] ?? 360;
     body = {
       model_id: state.model.id, scene_id: state.scene.id, roi: state.roi,
+      roi_poly: roiPoly,
       angle_min: 0, angle_max: angMax,
       angle_step: PREC_MAP[user.precisione] ?? 5,
       scale_min: smin, scale_max: smax, scale_step: sstep,
@@ -412,6 +521,7 @@ async function doMatch() {
   } else {
     body = {
       model_id: state.model.id, scene_id: state.scene.id, roi: state.roi,
+      roi_poly: roiPoly,
       ...user,
     };
   }
@@ -426,6 +536,12 @@ async function doMatch() {
   }
   const data = await r.json();
   state.matches = data.matches;
+  // C: salva tutto per "Esporta JSON"
+  state.lastMatch = {
+    endpoint: url, params: body, response: data,
+    image_id: state.scene.id,
+  };
+  document.getElementById("btn-export-json").disabled = false;
   state.annotatedImg = await loadImage(
     `/image/${data.annotated_id}/raw?t=${Date.now()}`);
   renderScene();
@@ -461,6 +577,38 @@ function setStatus(s) {
   document.getElementById("status").textContent = s;
 }
 
+// ---------- C: Export JSON risultati ----------
+function exportMatchJSON() {
+  if (!state.lastMatch) {
+    alert("Nessun match da esportare: esegui prima un MATCH.");
+    return;
+  }
+  const lm = state.lastMatch;
+  const payload = {
+    timestamp: new Date().toISOString(),
+    image_id: lm.image_id,
+    endpoint: lm.endpoint,
+    params: lm.params,
+    matches: lm.response.matches.map((m) => ({
+      cx: m.cx, cy: m.cy, angle_deg: m.angle_deg,
+      scale: m.scale, score: m.score, bbox: m.bbox_poly,
+    })),
+    train_time: lm.response.train_time,
+    find_time: lm.response.find_time,
+    num_variants: lm.response.num_variants,
+  };
+  const blob = new Blob([JSON.stringify(payload, null, 2)],
+                        { type: "application/json" });
+  const a = document.createElement("a");
+  a.href = URL.createObjectURL(blob);
+  const ts = new Date().toISOString().replace(/[:.]/g, "-");
+  a.download = `pm2d_match_${ts}.json`;
+  document.body.appendChild(a);
+  a.click();
+  a.remove();
+  URL.revokeObjectURL(a.href);
+}
+
 // ---------- Init ----------
 // ---------- Edge preview (clean rumore) ----------
 let _epDebounce = null;
@@ -734,6 +882,7 @@ async function saveRecipe() {
     model_id: state.model.id,
     scene_id: state.scene?.id || state.model.id,
     roi: state.roi,
+    roi_poly: getRoiPoly(),
     tipo: user.tipo,
     simmetria: user.simmetria,
     scala: user.scala,
@@ -777,6 +926,24 @@ window.addEventListener("DOMContentLoaded", async () => {
   upEl.addEventListener("change", async (e) => {
     const f = e.target.files[0];
     if (!f) return;
+    // A: file DXF → rasterizza server-side e usa direttamente come modello
+    if (f.name.toLowerCase().endsWith(".dxf")) {
+      setStatus(`Rasterizzazione DXF ${f.name}...`);
+      try {
+        const meta = await uploadDxf(f);
+        const img = await loadImage(`/image/${meta.id}/raw`);
+        state.model = { id: meta.id, w: meta.width, h: meta.height, img };
+        state.roi = null;
+        resetPoly();
+        setStatus(`DXF ${f.name} rasterizzato ` +
+                  `${meta.width}x${meta.height} — disegna ROI sul modello`);
+        renderModel();
+      } catch (err) {
+        setStatus(`Errore DXF: ${err.message}`);
+      }
+      e.target.value = "";
+      return;
+    }
     setStatus(`Caricamento ${f.name} nella cartella...`);
     try {
       const res = await uploadToFolder(f);
@@ -789,6 +956,18 @@ window.addEventListener("DOMContentLoaded", async () => {
   });
   document.getElementById("btn-match").addEventListener("click", doMatch);
   document.getElementById("btn-autotune").addEventListener("click", doAutoTune);
+  // B: ROI poligonale (toggle + chiudi + reset)
+  document.getElementById("roi-poly-toggle").addEventListener("change", (e) => {
+    state.polyMode = e.target.checked;
+    document.getElementById("btn-poly-close").disabled = !state.polyMode;
+    document.getElementById("btn-poly-reset").disabled = !state.polyMode;
+    resetPoly();
+  });
+  document.getElementById("btn-poly-close").addEventListener("click", closePoly);
+  document.getElementById("btn-poly-reset").addEventListener("click", resetPoly);
+  // C: export JSON ultimo match
+  document.getElementById("btn-export-json").addEventListener("click",
+    exportMatchJSON);
   document.getElementById("btn-save-recipe").addEventListener("click",
     saveRecipe);
   document.getElementById("btn-load-recipe").addEventListener("click",

pm2d/web/static/index.html

@@ -30,9 +30,9 @@
             title="Analizza ROI e derivata parametri ottimali (Halcon-style)">
       ⚙ Auto-tune
     </button>
-    <label class="btn" title="Carica nuovo file nella cartella immagini">
+    <label class="btn" title="Carica nuovo file nella cartella immagini (immagine o DXF)">
       ⬆ Carica file
-      <input type="file" id="file-upload" accept="image/*" hidden>
+      <input type="file" id="file-upload" accept="image/*,.dxf" hidden>
     </label>
     <span id="status">Seleziona modello, disegna ROI, seleziona scena</span>
   </div>
@@ -45,6 +45,15 @@
       <canvas id="c-model" width="380" height="420"></canvas>
     </div>
     <div id="roi-info">ROI: (nessuna)</div>
+    <div id="roi-poly-bar" style="display:flex; gap:6px; align-items:center; margin-top:6px">
+      <label style="display:flex; gap:4px; align-items:center; font-size:12px; cursor:pointer">
+        <input type="checkbox" id="roi-poly-toggle"> ROI poligonale
+      </label>
+      <button class="btn" id="btn-poly-close" type="button" disabled
+              title="Chiude il poligono (equivale al doppio click)">Chiudi</button>
+      <button class="btn" id="btn-poly-reset" type="button" disabled
+              title="Cancella i vertici del poligono">Reset</button>
+    </div>
     <details id="edge-preview-panel" style="margin-top:10px">
       <summary>🔬 Anteprima edge / pulizia rumore</summary>
       <div style="font-size:11px; color:#aaa; margin:4px 0">
@@ -248,6 +257,11 @@
     <div class="kv"><span>find:</span><span id="t-find">-</span></div>
     <div class="kv"><span>varianti:</span><span id="t-var">-</span></div>
     <div class="kv"><span>match:</span><span id="t-match">-</span></div>
+    <button class="btn" id="btn-export-json" type="button" disabled
+            style="margin-top:8px; width:100%"
+            title="Scarica i risultati dell'ultimo match in formato JSON">
+      ⬇ Esporta JSON
+    </button>
 
     <details id="diag-panel" style="margin-top:10px">
       <summary>🔍 Diagnostica (CC)</summary>

pyproject.toml

@@ -10,6 +10,7 @@ dependencies = [
     "pillow>=12.2.0",
     "python-multipart>=0.0.26",
     "uvicorn[standard]>=0.34",
+    "ezdxf>=1.3",
 ]
 
 [project.scripts]
@@ -19,4 +20,14 @@ pm2d-bench = "pm2d.bench:main"
 [dependency-groups]
 dev = [
     "httpx>=0.28.1",
+    "pytest>=8.0",
+    "ruff>=0.8",
 ]
+
+[tool.ruff]
+line-length = 100
+
+[tool.ruff.lint]
+select = ["E", "F"]
+# E702 (a; b) ed E402 (import dopo codice) sono idiomi voluti del codebase
+ignore = ["E501", "E741", "E702", "E731", "E402"]

tests/conftest.py

@@ -0,0 +1,99 @@
+"""Fixture condivise: template e scene sintetiche con ground-truth nota.
+
+Tutti i test sono sintetici (nessuna dipendenza dalle immagini Test/,
+non versionate): generano scene con pose note e verificano recall e
+precisione del matcher. Runtime totale atteso: ~2-4 min su 2 core.
+"""
+from __future__ import annotations
+
+import math
+
+import cv2
+import numpy as np
+import pytest
+
+
+def make_template(tw: int = 160, th: int = 120) -> np.ndarray:
+    """Forma a L asimmetrica con foro circolare, contrasto netto.
+
+    Asimmetrica per evitare ambiguita' rotazionali nei confronti GT.
+    """
+    img = np.full((th, tw), 60, np.uint8)
+    cv2.rectangle(img, (20, 20), (60, th - 20), 200, -1)
+    cv2.rectangle(img, (20, th - 55), (tw - 25, th - 20), 200, -1)
+    cv2.circle(img, (tw - 45, 40), 16, 200, -1)
+    return cv2.GaussianBlur(img, (3, 3), 0)
+
+
+# Pose ground-truth: (cx, cy, angle_deg) - angoli volutamente lontani
+# dalla griglia di step 5/2 gradi per misurare il refine.
+GT_POSES: list[tuple[float, float, float]] = [
+    (150.0, 150.0, 0.0),
+    (450.0, 140.0, 7.3),
+    (740.0, 170.0, 33.7),
+    (160.0, 420.0, 91.2),
+    (460.0, 430.0, 158.4),
+    (750.0, 480.0, 246.9),
+    (300.0, 590.0, 312.6),
+]
+
+
+def make_scene(
+    template: np.ndarray,
+    poses: list[tuple[float, float, float]],
+    W: int = 900, H: int = 700,
+    noise: float = 4.0, seed: int = 7,
+) -> np.ndarray:
+    """Incolla il template warpato alle pose date su sfondo rumoroso.
+
+    Convenzione di rotazione identica al matcher (cv2.getRotationMatrix2D
+    attorno al centro template, poi traslazione del centro su (cx, cy)).
+    """
+    rng = np.random.default_rng(seed)
+    scene = np.full((H, W), 60, np.float32)
+    th, tw = template.shape
+    for (cx, cy, ang) in poses:
+        M = cv2.getRotationMatrix2D((tw / 2.0, th / 2.0), ang, 1.0)
+        M[0, 2] += cx - tw / 2.0
+        M[1, 2] += cy - th / 2.0
+        warped = cv2.warpAffine(template.astype(np.float32), M, (W, H),
+                                flags=cv2.INTER_LINEAR, borderValue=-1)
+        scene = np.where(warped >= 0, warped, scene)
+    scene += rng.normal(0, noise, scene.shape)
+    return np.clip(scene, 0, 255).astype(np.uint8)
+
+
+def ang_diff(a: float, b: float) -> float:
+    """Differenza angolare firmata in (-180, 180]."""
+    d = (a - b) % 360.0
+    return d - 360.0 if d > 180.0 else d
+
+
+def match_errors(matches, poses, radius: float = 20.0):
+    """Associa match a pose GT per distanza; ritorna (err_ang, err_pos, n_miss)."""
+    errs_a: list[float] = []
+    errs_p: list[float] = []
+    miss = 0
+    for (cx, cy, ang) in poses:
+        cands = [
+            (math.hypot(m.cx - cx, m.cy - cy), m)
+            for m in matches
+            if math.hypot(m.cx - cx, m.cy - cy) < radius
+        ]
+        if not cands:
+            miss += 1
+            continue
+        d, m = min(cands, key=lambda t: t[0])
+        errs_a.append(abs(ang_diff(m.angle_deg, ang)))
+        errs_p.append(d)
+    return errs_a, errs_p, miss
+
+
+@pytest.fixture(scope="session")
+def template() -> np.ndarray:
+    return make_template()
+
+
+@pytest.fixture(scope="session")
+def scene(template) -> np.ndarray:
+    return make_scene(template, GT_POSES)

tests/test_matcher_unit.py

@@ -0,0 +1,84 @@
+"""Unit test rapidi su componenti del matcher (no matching pesante)."""
+from __future__ import annotations
+
+import numpy as np
+import cv2
+import pytest
+
+from pm2d import LineShapeMatcher
+from tests.conftest import GT_POSES, make_scene, match_errors
+
+
+def test_angle_list_includes_range_end():
+    # Range parziale ±15: l'estremo +15 deve essere testato (era escluso).
+    m = LineShapeMatcher(angle_range_deg=(-15.0, 15.0), angle_step_deg=5.0)
+    angles = m._angle_list()
+    assert -15.0 in angles and 15.0 in angles
+    assert len(angles) == 7
+
+
+def test_angle_list_full_circle_no_duplicate():
+    # (0, 360): 360 coincide con 0 → escluso, niente variante duplicata.
+    m = LineShapeMatcher(angle_range_deg=(0.0, 360.0), angle_step_deg=5.0)
+    angles = m._angle_list()
+    assert len(angles) == 72
+    assert 360.0 not in angles
+
+
+def test_pyramid_clamp_small_template():
+    # Template 40px di lato minimo: al top /4 le feature collassano →
+    # i livelli vengono clampati (40/2=20 >= 12, 40/4=10 < 12 → 2 livelli).
+    m = LineShapeMatcher(pyramid_levels=4, angle_range_deg=(0.0, 10.0),
+                         angle_step_deg=5.0)
+    tpl = np.full((40, 200), 60, np.uint8)
+    cv2.rectangle(tpl, (30, 8), (170, 32), 200, -1)
+    m.train(tpl)
+    assert m.pyramid_levels == 2
+
+
+def test_save_load_roundtrip(tmp_path, template, scene):
+    m = LineShapeMatcher(angle_step_deg=10.0)
+    m.train(template)
+    path = str(tmp_path / "model.npz")
+    m.save_model(path)
+    m2 = LineShapeMatcher.load_model(path)
+    assert len(m2.variants) == len(m.variants)
+    matches = m2.find(scene, min_score=0.5, max_matches=10)
+    _, _, miss = match_errors(matches, GT_POSES)
+    assert miss == 0
+
+
+def test_scene_cache_no_collision(template):
+    # Due scene IDENTICHE nella banda superiore ma diverse sotto: la cache
+    # (che prima hashava solo i primi 64KB) non deve restituire i risultati
+    # della scena sbagliata.
+    poses_a = [GT_POSES[0], (450.0, 560.0, 33.7)]
+    poses_b = [GT_POSES[0], (700.0, 560.0, 91.2)]
+    scene_a = make_scene(template, poses_a)
+    scene_b = make_scene(template, poses_b)
+    # Stessa banda superiore (le pose extra sono in basso, y >= 430)
+    assert np.array_equal(scene_a[:80], scene_b[:80])
+    m = LineShapeMatcher(angle_step_deg=10.0)
+    m.train(template)
+    ma = m.find(scene_a, min_score=0.5, max_matches=5)
+    mb = m.find(scene_b, min_score=0.5, max_matches=5)
+    _, _, miss_a = match_errors(ma, poses_a)
+    _, _, miss_b = match_errors(mb, poses_b)
+    assert miss_a == 0 and miss_b == 0
+
+
+def test_train_mask_polygonal(template, scene):
+    # ROI poligonale: mask che copre solo la L verticale del template.
+    mask = np.zeros_like(template)
+    cv2.rectangle(mask, (10, 10), (70, template.shape[0] - 10), 255, -1)
+    m = LineShapeMatcher(angle_step_deg=10.0)
+    n = m.train(template, mask=mask)
+    assert n > 0
+    matches = m.find(scene, min_score=0.5, max_matches=10)
+    assert len(matches) >= 1
+
+
+def test_untrained_find_raises():
+    m = LineShapeMatcher()
+    with pytest.raises(RuntimeError):
+        m.find(np.zeros((100, 100), np.uint8))

tests/test_precision.py

@@ -0,0 +1,56 @@
+"""Test di non-regressione su precisione e recall (GT sintetica).
+
+Soglie derivate dalle misure di Fase 2 (errore mediano ~0.05 deg /
+~0.08 px) con margine 3-4x per assorbire rumore tra run/macchine.
+Una regressione del refine (es. score saturo, minMaxLoc sul plateau)
+riporterebbe gli errori a 2-4 deg / 4 px e fa fallire i test con
+margine enorme.
+"""
+from __future__ import annotations
+
+import numpy as np
+
+from pm2d import LineShapeMatcher
+from tests.conftest import GT_POSES, match_errors
+
+
+def _find(template, scene, step, **kw):
+    m = LineShapeMatcher(angle_step_deg=step, num_features=96)
+    m.train(template)
+    return m.find(scene, min_score=0.5, max_matches=10, **kw)
+
+
+def test_recall_and_precision_step5(template, scene):
+    matches = _find(template, scene, 5.0)
+    errs_a, errs_p, miss = match_errors(matches, GT_POSES)
+    assert miss == 0, f"{miss} pose GT non trovate"
+    assert float(np.median(errs_a)) < 0.2, f"err angolo mediano {np.median(errs_a):.3f} deg"
+    assert float(np.max(errs_a)) < 0.5, f"err angolo max {np.max(errs_a):.3f} deg"
+    assert float(np.median(errs_p)) < 0.3, f"err posizione mediano {np.median(errs_p):.3f} px"
+    assert float(np.max(errs_p)) < 1.0, f"err posizione max {np.max(errs_p):.3f} px"
+
+
+def test_recall_and_precision_step2(template, scene):
+    # Step fine: storicamente il caso peggiore (plateau con piu' varianti
+    # dentro la tolleranza spread → scelta variante arbitraria).
+    matches = _find(template, scene, 2.0)
+    errs_a, errs_p, miss = match_errors(matches, GT_POSES)
+    assert miss == 0, f"{miss} pose GT non trovate"
+    assert float(np.median(errs_a)) < 0.2
+    assert float(np.max(errs_a)) < 0.5
+    assert float(np.median(errs_p)) < 0.3
+
+
+def test_no_false_positives(template, scene):
+    # max_matches alto: non devono comparire match spuri oltre le 7 pose.
+    matches = _find(template, scene, 5.0)
+    assert len(matches) <= len(GT_POSES) + 1, (
+        f"{len(matches)} match per {len(GT_POSES)} oggetti reali"
+    )
+
+
+def test_full_scan_path_equivalent(template, scene):
+    # Il path full-scan (propagate off) deve trovare le stesse pose.
+    matches = _find(template, scene, 5.0, pyramid_propagate=False)
+    _, _, miss = match_errors(matches, GT_POSES)
+    assert miss == 0

uv.lock

@@ -62,6 +62,29 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/d1/d6/3965ed04c63042e047cb6a3e6ed1a63a35087b6a609aa3a15ed8ac56c221/colorama-0.4.6-py2.py3-none-any.whl", hash = "sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6", size = 25335, upload-time = "2022-10-25T02:36:20.889Z" },
 ]
 
+[[package]]
+name = "ezdxf"
+version = "1.4.4"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "fonttools" },
+    { name = "numpy" },
+    { name = "pyparsing" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/5e/d7/1b7be8db364f1c4838dfc1a40ca96577aba405deabf896a4eb3aaeb15a62/ezdxf-1.4.4.tar.gz", hash = "sha256:da5a5e0e6bdbb6656f9c017b47edc7eafceb419d61a2b5de64ffb344c168e593", size = 1866886, upload-time = "2026-05-14T09:19:19.511Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/be/ed/97aa3ba1ba923e4098169de5bca380ee1430e2b34d0dd85b73007e34df16/ezdxf-1.4.4-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:278a09a845d67a4f893aebe9ef8dcd8894ece9f255d7d8bb21719ffd902109b8", size = 3555283, upload-time = "2026-05-14T09:25:37.346Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/b1/8ca9408a2b382a806837e72d35f10931fdd9f53fe66cf208a6340358fbc5/ezdxf-1.4.4-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:b0a787f8ebcbdcb9798c6fcfe90c23afd1a0993c8d4302b6cce5d8f19ae052ae", size = 3012646, upload-time = "2026-05-14T09:25:39.01Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/aa/1e8d3130eaabb780c5e8adfdc1ead023bb04468779b18169d60f7ce5b291/ezdxf-1.4.4-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:c451d1c20f08b30c735ca4d650fee5b44147402f4b5662f8550053ed8a3009c2", size = 2999048, upload-time = "2026-05-14T09:25:40.348Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/ca/1dd390b79df9c104c2d4b5964e80db25362edf704ac742c9db9b279658a5/ezdxf-1.4.4-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3d3d9a6663993fc644751647fd7843057b51742ffac27c48ce28ea9f81239613", size = 5747890, upload-time = "2026-05-14T09:27:36.028Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/1e/226a6e636ae533e0bc0afe4435d1058fac173564e4af4102862f055a46d9/ezdxf-1.4.4-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:82c022cf094d21ad3db68557aa83628e4152355e8a8bef0fffd4f71a9ea325df", size = 5774707, upload-time = "2026-05-14T09:27:17.625Z" },
+    { url = "https://files.pythonhosted.org/packages/24/a3/7e33c9036de944b5446982bb629d698356e48b00c4d173768f52bdeeff4b/ezdxf-1.4.4-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:3be59c2a3a93585412a2a37af29b3ad7574fa3e429e624924c3285d0573388b3", size = 5713753, upload-time = "2026-05-14T09:27:37.55Z" },
+    { url = "https://files.pythonhosted.org/packages/36/9e/ac0cdc3a8623fc38aa16670f6537aa810bc1381217a51ac3481299bbc986/ezdxf-1.4.4-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:44a9716860b59dddd49a3708c04ff4580ad1e7de92ac93bb3213d85fba5ac93f", size = 5813895, upload-time = "2026-05-14T09:27:19.514Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/22/b511a8d9ea8f23447a67fe8977213fac5e4e69df30130a3ef31043613eaf/ezdxf-1.4.4-cp313-cp313-win_amd64.whl", hash = "sha256:207eed544417464ffaf2570880d58a792ecd2534c9c2dede45e895f3200b77e6", size = 2960937, upload-time = "2026-05-14T09:23:49.361Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/09/5ecb6f82d35a2f4a4334d0a608fcf764827fa69dccdf5987ce309c16b6c1/ezdxf-1.4.4-py3-none-any.whl", hash = "sha256:666edda631ba717270293b734f5d58dd97a1d1aba4787187f09d0cc584645865", size = 1331217, upload-time = "2026-05-14T09:19:26.601Z" },
+]
+
 [[package]]
 name = "fastapi"
 version = "0.136.1"
@@ -78,6 +101,39 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/5a/ff/2e4eca3ade2c22fe1dea7043b8ee9dabe47753349eb1b56a202de8af6349/fastapi-0.136.1-py3-none-any.whl", hash = "sha256:a6e9d7eeada96c93a4d69cb03836b44fa34e2854accb7244a1ece36cd4781c3f", size = 117683, upload-time = "2026-04-23T16:49:42.437Z" },
 ]
 
+[[package]]
+name = "fonttools"
+version = "4.63.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/84/69/c97f2c18e0db87d2c7b15da1974dace76ae938f1cfa22e2727a648b7ed43/fonttools-4.63.0.tar.gz", hash = "sha256:caeb583deeb5168e694b65cda8b4ee62abedfa66cf88488734466f2366b9c4e0", size = 3597189, upload-time = "2026-05-14T12:04:30.958Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0f/8d/d8fec3dcde2963f8c908fb315e5ff2cd0ac34f82394bbbf73a2aa5145ce3/fonttools-4.63.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:cd7e9857e5e63738b9d9fd707bc1f59c8b09e5177726d23664db393c59bb08bd", size = 2876062, upload-time = "2026-05-14T12:03:32.554Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/71/d935dc54e4ff121bfdd11e08702db63a7e6f25af21d8a3d7b7212df53641/fonttools-4.63.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:c2a2a42198b696a6f48fad91709afb55176e66a5e566131219dba372fb7f8c59", size = 2424594, upload-time = "2026-05-14T12:03:34.86Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/40/e76320afa1df918e146155ef239b1719ee266092e96f5423bfd075affba1/fonttools-4.63.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:1e874792a8212b44583ea02189d9e693906b2f78b261f372f95d6c563210ac1d", size = 5024840, upload-time = "2026-05-14T12:03:36.745Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/36/0b805d8c485f872f65a509cbe3b58a5d0d17bee855333b54a150c79d3061/fonttools-4.63.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:22135da48a348785c5e2d5d2d9d6bec5ed44adacbaeb9db12d9493bf6c6bfa68", size = 4975801, upload-time = "2026-05-14T12:03:38.833Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/26/2cee03d0aa083ab022da5c07aff9ed3f689da1defb81ad6917c9627896da/fonttools-4.63.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:ccf41f2efdf56994d22d73bef4ced1052161958169428d06ba9724ea9e9a64be", size = 4965009, upload-time = "2026-05-14T12:03:41.494Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/48/cc4b66d9058c0d0982c833fad10127c4b0e9324606aafa41382295ca4102/fonttools-4.63.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:9ced0bd02ac751dd6319b0da88aaef24414e3b0dbc32bb4f24944821a3741a27", size = 5105892, upload-time = "2026-05-14T12:03:43.525Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/1f/a98a30a814b9ddef3a2e706025f90b9e0bc94890e6cb15254bc86547d11a/fonttools-4.63.0-cp313-cp313-win32.whl", hash = "sha256:85be818f5506e8a7753153def2c9550178f0ecae6a47b5e0e8dbb23f7cc90380", size = 2291313, upload-time = "2026-05-14T12:03:45.594Z" },
+    { url = "https://files.pythonhosted.org/packages/92/46/5177b01f3b4abfdd4409f31cca4ab279c9343a26efbe9ec78c97fc612e02/fonttools-4.63.0-cp313-cp313-win_amd64.whl", hash = "sha256:ba04cb5891d4c0c21b6da95eda8d7b090021508a294fff33464fc7d241e0856b", size = 2342299, upload-time = "2026-05-14T12:03:47.414Z" },
+    { url = "https://files.pythonhosted.org/packages/27/d2/23d25e3f247b328be58d04a4c9f894178a0d1eda7d42867cfb388adaf416/fonttools-4.63.0-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:fd1e3094f42d806d3d7c79162fc59e5910fcbe3a7360c385b8da969bc4493745", size = 2875338, upload-time = "2026-05-14T12:03:50.052Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/58/7dfa0c761cb3b2964e2a84c4dc986c926a87de0cb9fb60d5b28ded3f2914/fonttools-4.63.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:6e528da43bc3791085f8cb6141b1d13e459226790240340fcbb4625649238b03", size = 2422661, upload-time = "2026-05-14T12:03:52.154Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/87/64cfa18a7a1621d17b7f4502b2b0ed8a135a90c3db51ea590ee99043e76b/fonttools-4.63.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6b2248c5decb223562f7902ff6325077a073f608ee8e33e88ad88db734eb9f49", size = 5010526, upload-time = "2026-05-14T12:03:54.647Z" },
+    { url = "https://files.pythonhosted.org/packages/36/e1/a8933a72c45a87177fbde2696e0d0755c8c9062f8c077a961c6215fa27b1/fonttools-4.63.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:308f957cdeaf8abe4e5f2f124902ef405448af92c90f80e302a3b771c2e6116b", size = 4923946, upload-time = "2026-05-14T12:03:56.984Z" },
+    { url = "https://files.pythonhosted.org/packages/27/60/872e6e233b8c5e8b41413796ff18b7fe479661bd40147e071b450dfad7a1/fonttools-4.63.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:bf00f21eb5fb721dbaf73d1e9da6d02a1af7768f2ebcf9798be98beab8ba90f6", size = 4962489, upload-time = "2026-05-14T12:03:59.443Z" },
+    { url = "https://files.pythonhosted.org/packages/30/c4/83c24f2ec38b90cfda84bf4b1a1f49df80e84a1db4e7ac6e0d41bf23bc39/fonttools-4.63.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:c1aaa4b9c75798400ac043ce04d74e7830376c85095a5a6ed7cba2f17a266bf4", size = 5071870, upload-time = "2026-05-14T12:04:02.122Z" },
+    { url = "https://files.pythonhosted.org/packages/de/40/3ae22b60ff1d41ce0bd044b31238cdc72cef99f28b976f1e128ebd618c9b/fonttools-4.63.0-cp314-cp314-win32.whl", hash = "sha256:22693918177bd9ceabec4736d338045f357769416fc6b0b2508eefef75b08616", size = 2295026, upload-time = "2026-05-14T12:04:04.47Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/d4/98078064ccc76b45cb0f6c002452011e93c4bd26f6850344f0951cc1fe89/fonttools-4.63.0-cp314-cp314-win_amd64.whl", hash = "sha256:7d782fac32985914c351556f68ac0855391572bcd87de50e05970d3cd4c96fc5", size = 2347454, upload-time = "2026-05-14T12:04:06.752Z" },
+    { url = "https://files.pythonhosted.org/packages/49/4e/652d1580c5f4e39f7d103b0c793e4773129ad633dce4addd0cf4dfebde02/fonttools-4.63.0-cp314-cp314t-macosx_10_15_universal2.whl", hash = "sha256:6db5140a60a5d731d21ec076745b40a310607731b0a565b50776393188649001", size = 2958152, upload-time = "2026-05-14T12:04:08.706Z" },
+    { url = "https://files.pythonhosted.org/packages/0e/55/ad864c9a9b219f552eb46b32cd7906c466e5a578ba0c3abfcc0fe7413eb6/fonttools-4.63.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:7d76edbff9014094dbf03bd2d074709dfa6ec7aba13d838c937a2b33d2d6a86e", size = 2460809, upload-time = "2026-05-14T12:04:10.783Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/2b/0aa8db70f18cf52e49b4ed5ecec68547f981160bf5ded3b5aed6faa0a6f9/fonttools-4.63.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0eac00b9118c3c2f87d272e45341871c5b3066baa3c86897fa634a7c3fb59096", size = 5148649, upload-time = "2026-05-14T12:04:12.747Z" },
+    { url = "https://files.pythonhosted.org/packages/7f/63/18e4369c25043096f1048e0c9915951adc4f842bd81c6b18155824d6fa99/fonttools-4.63.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:51394295f1a51de8b5f30bdb1e1b9a4231536c7064ef5c6e211eec19fa36036f", size = 4932147, upload-time = "2026-05-14T12:04:14.806Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/3f/67f3eac2ffd8a98446c5022f8ed3864eac878a5ff7af8df4c8286dba16cc/fonttools-4.63.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:9e12f105d2b6342c559c298afb674006bb2893afc7102dcf8a1b55b0486b4e40", size = 5027237, upload-time = "2026-05-14T12:04:17.675Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/ba/4e6214cb38a7b04779e97bb7636de9a5c7f20af7018d03dee0b64c08510a/fonttools-4.63.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:796f27556dbe094c4824f75ca85267e4df776c79036c8441469a4df37038c196", size = 5053933, upload-time = "2026-05-14T12:04:20.818Z" },
+    { url = "https://files.pythonhosted.org/packages/34/3b/214dcc19ee31d3d38fb5ad2755c11ef0514e5dc300bbaf41c0b69f393799/fonttools-4.63.0-cp314-cp314t-win32.whl", hash = "sha256:948428a275741f0b64b113c955425a953314f4b9ab9997f73a72c83e68e569c8", size = 2359326, upload-time = "2026-05-14T12:04:24.22Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/1e/3ff1a9b523058c2eeb6a9d50f5574e2a738200d0d94107d5bc4105e8da3f/fonttools-4.63.0-cp314-cp314t-win_amd64.whl", hash = "sha256:6d4741eb179121cab9eea4cb2393d24492373a260d7945006358c08cfbf45419", size = 2425829, upload-time = "2026-05-14T12:04:26.829Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/47/c99d5268f354002ce80f8d029cd9d7d872969da1de8b93d32de4dc56d6f4/fonttools-4.63.0-py3-none-any.whl", hash = "sha256:445af2eab030a16b9171ea8bdda7ebf7d96bda2df88ee182a464252f6e05e20d", size = 1164562, upload-time = "2026-05-14T12:04:29.092Z" },
+]
+
 [[package]]
 name = "h11"
 version = "0.16.0"
@@ -146,6 +202,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/5d/13/ad7d7ca3808a898b4612b6fe93cde56b53f3034dcde235acb1f0e1df24c6/idna-3.13-py3-none-any.whl", hash = "sha256:892ea0cde124a99ce773decba204c5552b69c3c67ffd5f232eb7696135bc8bb3", size = 68629, upload-time = "2026-04-22T16:42:40.909Z" },
 ]
 
+[[package]]
+name = "iniconfig"
+version = "2.3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/72/34/14ca021ce8e5dfedc35312d08ba8bf51fdd999c576889fc2c24cb97f4f10/iniconfig-2.3.0.tar.gz", hash = "sha256:c76315c77db068650d49c5b56314774a7804df16fee4402c1f19d6d15d8c4730", size = 20503, upload-time = "2025-10-18T21:55:43.219Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/cb/b1/3846dd7f199d53cb17f49cba7e651e9ce294d8497c8c150530ed11865bb8/iniconfig-2.3.0-py3-none-any.whl", hash = "sha256:f631c04d2c48c52b84d0d0549c99ff3859c98df65b3101406327ecc7d53fbf12", size = 7484, upload-time = "2025-10-18T21:55:41.639Z" },
+]
+
 [[package]]
 name = "llvmlite"
 version = "0.47.0"
@@ -258,6 +323,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/e9/a5/1be1516390333ff9be3a9cb648c9f33df79d5096e5884b5df71a588af463/opencv_python-4.13.0.92-cp37-abi3-win_amd64.whl", hash = "sha256:423d934c9fafb91aad38edf26efb46da91ffbc05f3f59c4b0c72e699720706f5", size = 40212062, upload-time = "2026-02-05T07:02:12.724Z" },
 ]
 
+[[package]]
+name = "packaging"
+version = "26.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d7/f1/e7a6dd94a8d4a5626c03e4e99c87f241ba9e350cd9e6d75123f992427270/packaging-26.2.tar.gz", hash = "sha256:ff452ff5a3e828ce110190feff1178bb1f2ea2281fa2075aadb987c2fb221661", size = 228134, upload-time = "2026-04-24T20:15:23.917Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/df/b2/87e62e8c3e2f4b32e5fe99e0b86d576da1312593b39f47d8ceef365e95ed/packaging-26.2-py3-none-any.whl", hash = "sha256:5fc45236b9446107ff2415ce77c807cee2862cb6fac22b8a73826d0693b0980e", size = 100195, upload-time = "2026-04-24T20:15:22.081Z" },
+]
+
 [[package]]
 name = "pillow"
 version = "12.2.0"
@@ -316,6 +390,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/ff/6e/cf826fae916b8658848d7b9f38d88da6396895c676e8086fc0988073aaf8/pillow-12.2.0-cp314-cp314t-win_arm64.whl", hash = "sha256:aa88ccfe4e32d362816319ed727a004423aab09c5cea43c01a4b435643fa34eb", size = 2556579, upload-time = "2026-04-01T14:45:52.529Z" },
 ]
 
+[[package]]
+name = "pluggy"
+version = "1.6.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f9/e2/3e91f31a7d2b083fe6ef3fa267035b518369d9511ffab804f839851d2779/pluggy-1.6.0.tar.gz", hash = "sha256:7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3", size = 69412, upload-time = "2025-05-15T12:30:07.975Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl", hash = "sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746", size = 20538, upload-time = "2025-05-15T12:30:06.134Z" },
+]
+
 [[package]]
 name = "pydantic"
 version = "2.13.3"
@@ -387,6 +470,40 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/0b/db/d8182a7f1d9343a032265aae186eb063fe26ca4c40f256b21e8da4498e89/pydantic_core-2.46.3-cp314-cp314t-win_arm64.whl", hash = "sha256:77706aeb41df6a76568434701e0917da10692da28cb69d5fb6919ce5fdb07374", size = 2026310, upload-time = "2026-04-20T14:41:01.778Z" },
 ]
 
+[[package]]
+name = "pygments"
+version = "2.20.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c3/b2/bc9c9196916376152d655522fdcebac55e66de6603a76a02bca1b6414f6c/pygments-2.20.0.tar.gz", hash = "sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f", size = 4955991, upload-time = "2026-03-29T13:29:33.898Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f4/7e/a72dd26f3b0f4f2bf1dd8923c85f7ceb43172af56d63c7383eb62b332364/pygments-2.20.0-py3-none-any.whl", hash = "sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176", size = 1231151, upload-time = "2026-03-29T13:29:30.038Z" },
+]
+
+[[package]]
+name = "pyparsing"
+version = "3.3.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f3/91/9c6ee907786a473bf81c5f53cf703ba0957b23ab84c264080fb5a450416f/pyparsing-3.3.2.tar.gz", hash = "sha256:c777f4d763f140633dcb6d8a3eda953bf7a214dc4eff598413c070bcdc117cbc", size = 6851574, upload-time = "2026-01-21T03:57:59.36Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/10/bd/c038d7cc38edc1aa5bf91ab8068b63d4308c66c4c8bb3cbba7dfbc049f9c/pyparsing-3.3.2-py3-none-any.whl", hash = "sha256:850ba148bd908d7e2411587e247a1e4f0327839c40e2e5e6d05a007ecc69911d", size = 122781, upload-time = "2026-01-21T03:57:55.912Z" },
+]
+
+[[package]]
+name = "pytest"
+version = "9.0.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "iniconfig" },
+    { name = "packaging" },
+    { name = "pluggy" },
+    { name = "pygments" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/7d/0d/549bd94f1a0a402dc8cf64563a117c0f3765662e2e668477624baeec44d5/pytest-9.0.3.tar.gz", hash = "sha256:b86ada508af81d19edeb213c681b1d48246c1a91d304c6c81a427674c17eb91c", size = 1572165, upload-time = "2026-04-07T17:16:18.027Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d4/24/a372aaf5c9b7208e7112038812994107bc65a84cd00e0354a88c2c77a617/pytest-9.0.3-py3-none-any.whl", hash = "sha256:2c5efc453d45394fdd706ade797c0a81091eccd1d6e4bccfcd476e2b8e0ab5d9", size = 375249, upload-time = "2026-04-07T17:16:16.13Z" },
+]
+
 [[package]]
 name = "python-dotenv"
 version = "1.2.2"
@@ -441,11 +558,37 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/f1/12/de94a39c2ef588c7e6455cfbe7343d3b2dc9d6b6b2f40c4c6565744c873d/pyyaml-6.0.3-cp314-cp314t-win_arm64.whl", hash = "sha256:ebc55a14a21cb14062aa4162f906cd962b28e2e9ea38f9b4391244cd8de4ae0b", size = 149341, upload-time = "2025-09-25T21:32:56.828Z" },
 ]
 
+[[package]]
+name = "ruff"
+version = "0.15.17"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/8c/a9/3abdf488f1bf3d24c699415e454ed554a6350d5d89ce183be1ee0a3361ac/ruff-0.15.17.tar.gz", hash = "sha256:2ec446937fd16c8c4de2674a209cc5af64d9c6f17d21fbf1151054fa0bcf5219", size = 4743346, upload-time = "2026-06-11T17:54:47.663Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/db/4d/e11259f5da07cb6afb2d074c31bf09da9671993f7329d4f15d2fdc458301/ruff-0.15.17-py3-none-linux_armv6l.whl", hash = "sha256:d9feddb927fc68bd295f5eebc587a7e42cfaf9b65f60ca4a2386febff575da8f", size = 10856677, upload-time = "2026-06-11T17:54:49.533Z" },
+    { url = "https://files.pythonhosted.org/packages/29/3e/772d679e1a0dc058e58875bd2c0cb713a0530877b4a76fee3c7966df0d49/ruff-0.15.17-py3-none-macosx_10_12_x86_64.whl", hash = "sha256:25805a226d741c47d274a35ad5c10a7dde175fcddfa511d7cf3da0a21eb3eab7", size = 11223443, upload-time = "2026-06-11T17:55:00.573Z" },
+    { url = "https://files.pythonhosted.org/packages/68/58/bd41f7688b2fd5623012605130ed70e60aa7f2244baa3d5066bdd61530c8/ruff-0.15.17-py3-none-macosx_11_0_arm64.whl", hash = "sha256:f6ad73b14c2d18a3bf8ad7cb6974294d7f613a7898604826058e6ac64918ef4d", size = 10566458, upload-time = "2026-06-11T17:55:07.52Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/5b/733371013fcf1ec339e477ece6ab42bfe10bdd9bba8ee88a9516aa56bfc0/ruff-0.15.17-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6ba0c1e4f95bcb3869d0d30cbd5917071ef2e28665abfec970cdab0492c713ed", size = 10914483, upload-time = "2026-06-11T17:55:05.501Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/cc/6f24251cc0252f7239391ccb85833f320efad14ebe5b443943f37ced6332/ruff-0.15.17-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:81647960f10bff57d2e51cadd0c3950fe598400c852863a038720ef5b8cca91e", size = 10647497, upload-time = "2026-06-11T17:54:57.733Z" },
+    { url = "https://files.pythonhosted.org/packages/68/dd/0d10c17ce1a1624d6fc3156309c3f834fdb5dfaad026ec90c85684f3990e/ruff-0.15.17-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0e01a84ddbc8c16c23055ba3924476850f1bbc1917cebbb9376665a63e74260d", size = 11416967, upload-time = "2026-06-11T17:54:51.461Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/91/556bfb156f6144f355e831c23db00b2fc4120f86b3ce81cc5f7fd2df51f3/ruff-0.15.17-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:84fe9f653152f8f294f9f7e03bf3a453d8b4a27f7a59c78c8666167f2b17b96c", size = 12335770, upload-time = "2026-06-11T17:54:45.793Z" },
+    { url = "https://files.pythonhosted.org/packages/88/82/8b5999aa13355e926f06d9f42a32dcca862f623bf0363785ff89d607dffd/ruff-0.15.17-py3-none-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:8c0fe88a7676e7a05b73174d4d4a59cb2ac21ff8263583f87a81a6018475a978", size = 11575441, upload-time = "2026-06-11T17:54:32.661Z" },
+    { url = "https://files.pythonhosted.org/packages/11/93/f10377bb04109ca0e8cbc483ff1982c54b6d418210041776f93e8cdc7fa9/ruff-0.15.17-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ecfc3c7878fff94633ab0348524e093f9ce3243080416dd7d14f8ba400174719", size = 11557614, upload-time = "2026-06-11T17:54:34.698Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/a6/eeeae7f7d5493df41649ab3db92f086b2d0a30199e4efdf8e3dd7a033f24/ruff-0.15.17-py3-none-manylinux_2_31_riscv64.whl", hash = "sha256:b8461180b22420b1bdc289909410930761629fddf2a5aaf60fae1ab26cedc4c4", size = 11544450, upload-time = "2026-06-11T17:54:39.042Z" },
+    { url = "https://files.pythonhosted.org/packages/32/88/5991ce565129a24dd4a00db1254b3b5db2e53018cbe4018ea5a89738e727/ruff-0.15.17-py3-none-musllinux_1_2_aarch64.whl", hash = "sha256:6eccbe50a038b503e7140b441aa9c7fc8c1f36edf23ebef9f4165c2f28f568b7", size = 10892524, upload-time = "2026-06-11T17:55:09.432Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/1d/0fdd248313425f55223968af04b0a42125466a8d88d21c1d99c6af0a51e8/ruff-0.15.17-py3-none-musllinux_1_2_armv7l.whl", hash = "sha256:382fc0521025f5a8ad447d8bdd523545d0d7646adb718eb1c2dac5065ec27c0f", size = 10659573, upload-time = "2026-06-11T17:54:36.824Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/0e/072e8260deb9461062ce9311ced27a8e541229a6ffd483013dd37661e43e/ruff-0.15.17-py3-none-musllinux_1_2_i686.whl", hash = "sha256:456d41fcd1b2777ad63f09a6e7121d43f7b688bbc76a800c10f7f8fb1f912c3f", size = 11127818, upload-time = "2026-06-11T17:55:03.124Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/b4/55060a34163121498014696b5f656db5b8c6963768f227dbf0d76b311073/ruff-0.15.17-py3-none-musllinux_1_2_x86_64.whl", hash = "sha256:b1a04bcc94ae6194e9db05d16ad31f298a7194bfbcb08258bbe589cee1d587b8", size = 11655901, upload-time = "2026-06-11T17:54:53.562Z" },
+    { url = "https://files.pythonhosted.org/packages/49/71/9b29d6b87cef468d697f43c6a91e3fae4a80185779d7d5a4ef27d173439f/ruff-0.15.17-py3-none-win32.whl", hash = "sha256:596065960ab1ff593f744220c9fe6580eda00a95003cffa9f4048bb5b1bf0392", size = 10925574, upload-time = "2026-06-11T17:54:55.723Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/b2/8fc77f3723228836fa5d12497eb71c808f83782e10d058d2b15cfa14640b/ruff-0.15.17-py3-none-win_amd64.whl", hash = "sha256:6769e5fa1710b179b92e0bfa5a51735b35baea9013dadb06d5f44cbcf9547084", size = 12058788, upload-time = "2026-06-11T17:54:41.042Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/c7/c53e8dbff9c9dc4b7928773421ae294a5d28fcb8dcda1a089579d3a7e510/ruff-0.15.17-py3-none-win_arm64.whl", hash = "sha256:f3be1fbb34bcdfd146240d8fb92a709d4c2c8191348580a3c044ec60fa0b4456", size = 11355275, upload-time = "2026-06-11T17:54:43.635Z" },
+]
+
 [[package]]
 name = "shape-model-2d"
 version = "0.1.0"
 source = { virtual = "." }
 dependencies = [
+    { name = "ezdxf" },
     { name = "fastapi" },
     { name = "numba" },
     { name = "numpy" },
@@ -458,10 +601,13 @@ dependencies = [
 [package.dev-dependencies]
 dev = [
     { name = "httpx" },
+    { name = "pytest" },
+    { name = "ruff" },
 ]
 
 [package.metadata]
 requires-dist = [
+    { name = "ezdxf", specifier = ">=1.3" },
     { name = "fastapi", specifier = ">=0.115" },
     { name = "numba", specifier = ">=0.65.0" },
     { name = "numpy", specifier = ">=1.24" },
@@ -472,7 +618,11 @@ requires-dist = [
 ]
 
 [package.metadata.requires-dev]
-dev = [{ name = "httpx", specifier = ">=0.28.1" }]
+dev = [
+    { name = "httpx", specifier = ">=0.28.1" },
+    { name = "pytest", specifier = ">=8.0" },
+    { name = "ruff", specifier = ">=0.8" },
+]
 
 [[package]]
 name = "starlette"