update OCR

src/inference.py

@@ -1,3 +1,12 @@
+# ================== FIX LỖI PYTHON 3.12 + PADDLEOCR ==================
+import os
+os.environ["PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK"] = "True"
+os.environ["PYTHONASYNCIODEBUG"] = "0"
+
+import warnings
+warnings.filterwarnings("ignore", category=UserWarning)
+# =====================================================================
+
 # src/inference.py
 import torch
 _orig_torch_load = torch.load
@@ -140,17 +149,25 @@ def ocr_table(img_path):
         # Fallback về OCR thường
         fallback_text = ocr_note(img_path)
         return {"rows": [], "text": fallback_text}
-def run_pipeline(image_path, output_dir="outputs",
-                 checkpoint="best.pt", conf_thresh=0.3):
+def run_pipeline(image_path, output_dir="outputs", checkpoint="best.pt", conf_thresh=0.25):
     image_path = str(image_path)
-    img_name   = Path(image_path).name
-    stem       = Path(image_path).stem
-    crop_dir   = Path(output_dir) / stem / "crops"
+    img_name = Path(image_path).name
+    stem = Path(image_path).stem
+    crop_dir = Path(output_dir) / stem / "crops"
     crop_dir.mkdir(parents=True, exist_ok=True)
 
-    model   = get_det_model(checkpoint)
-    results = model(image_path, imgsz=1024, conf=conf_thresh,
-                    iou=0.5, device=DEVICE, verbose=False)
+    model = get_det_model(checkpoint)
+    
+    # === DEBUG: In ra tất cả box mà model detect được ===
+    print(f"\n🔍 [DEBUG] Running detection on {img_name} (conf={conf_thresh})")
+    results = model(image_path, imgsz=1024, conf=conf_thresh, iou=0.5, device=DEVICE, verbose=False)
+    
+    print(f"   → Tổng số box detect được: {len(results[0].boxes)}")
+    for i, box in enumerate(results[0].boxes):
+        cls_idx = int(box.cls[0])
+        cls_name_raw = CLASS_NAMES[cls_idx] if cls_idx < len(CLASS_NAMES) else f"unknown_{cls_idx}"
+        conf = float(box.conf[0])
+        print(f"   Box {i+1}: class = '{cls_name_raw}' | conf = {conf:.3f}")
 
     img_bgr = cv2.imread(image_path)
     if img_bgr is None:
@@ -158,15 +175,18 @@ def run_pipeline(image_path, output_dir="outputs",
 
     objects = []
     for i, box in enumerate(results[0].boxes):
-        x1,y1,x2,y2 = map(int, box.xyxy[0].tolist())
-        cls_idx  = int(box.cls[0])
+        x1, y1, x2, y2 = map(int, box.xyxy[0].tolist())
+        cls_idx = int(box.cls[0])
         conf_val = round(float(box.conf[0]), 4)
-        cls_raw  = CLASS_NAMES[cls_idx]
-        cls_show = CLASS_DISPLAY[cls_raw]
+        
+        # Map class an toàn
+        cls_raw = CLASS_NAMES[cls_idx] if cls_idx < len(CLASS_NAMES) else "unknown"
+        cls_show = CLASS_DISPLAY.get(cls_raw, cls_raw.capitalize())
 
-        pad  = 6
-        crop = img_bgr[max(0,y1-pad):min(img_bgr.shape[0],y2+pad),
-                       max(0,x1-pad):min(img_bgr.shape[1],x2+pad)]
+        pad = 6
+        crop = img_bgr[max(0, y1-pad):min(img_bgr.shape[0], y2+pad),
+                       max(0, x1-pad):min(img_bgr.shape[1], x2+pad)]
+        
         crop_path = str(crop_dir / f"{cls_show}_{i+1}.jpg")
         cv2.imwrite(crop_path, crop, [cv2.IMWRITE_JPEG_QUALITY, 95])
 
@@ -174,38 +194,38 @@ def run_pipeline(image_path, output_dir="outputs",
         if cls_raw == "note":
             print(f"[OCR] Note #{i+1}...")
             ocr_content = ocr_note(crop_path)
-            print(f"      → {repr(ocr_content[:80]) if ocr_content else 'EMPTY'}")
         elif cls_raw == "table":
             print(f"[OCR] Table #{i+1}...")
             ocr_content = ocr_table(crop_path)
-            preview = ocr_content.get("text","")[:80]
-            print(f"      → {repr(preview) if preview else 'EMPTY'}")
 
         objects.append({
-            "id": i+1, "class": cls_show,
+            "id": i + 1,
+            "class": cls_show,
             "confidence": conf_val,
-            "bbox": {"x1":x1,"y1":y1,"x2":x2,"y2":y2},
+            "bbox": {"x1": x1, "y1": y1, "x2": x2, "y2": y2},
             "crop_path": crop_path,
             "ocr_content": ocr_content,
         })
 
-        color = COLORS[cls_raw]
-        cv2.rectangle(img_bgr, (x1,y1), (x2,y2), color, 2)
+        # Vẽ box
+        color = COLORS.get(cls_raw, (255, 255, 255))
+        cv2.rectangle(img_bgr, (x1, y1), (x2, y2), color, 2)
         label = f"{cls_show} {conf_val:.2f}"
-        (tw,th),_ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.6, 2)
-        cv2.rectangle(img_bgr, (x1,y1-th-10), (x1+tw+8,y1), color, -1)
-        cv2.putText(img_bgr, label, (x1+4,y1-4),
-                    cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255,255,255), 2)
+        (tw, th), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.6, 2)
+        cv2.rectangle(img_bgr, (x1, y1 - th - 10), (x1 + tw + 8, y1), color, -1)
+        cv2.putText(img_bgr, label, (x1 + 4, y1 - 4),
+                    cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255), 2)
 
-    vis_path = str(Path(output_dir)/stem/"result_vis.jpg")
+    # Lưu visualization và JSON
+    vis_path = str(Path(output_dir) / stem / "result_vis.jpg")
     cv2.imwrite(vis_path, img_bgr)
 
-    result    = {"image": img_name, "objects": objects}
-    json_path = str(Path(output_dir)/stem/"result.json")
-    with open(json_path,"w",encoding="utf-8") as f:
+    result = {"image": img_name, "objects": objects}
+    json_path = str(Path(output_dir) / stem / "result.json")
+    with open(json_path, "w", encoding="utf-8") as f:
         json.dump(result, f, ensure_ascii=False, indent=2)
 
-    print(f"[✓] {len(objects)} objects | {vis_path} | {json_path}")
+    print(f"\n[✓] Hoàn thành: {len(objects)} objects | JSON: {json_path} | Vis: {vis_path}")
     return result, vis_path
 
 if __name__ == "__main__":