Upload app.py

app.py

@@ -137,7 +137,7 @@ CONFIGS = {
         'model_path': 'best_upright.onnx',
         'yaml_path': 'data_upright.yaml',
         'input_size': 640,
-        'confidence_threshold': 0.45,
+        'confidence_threshold': 0.25,  # Lowered from 0.45 for better detection
         'nms_threshold': 0.45
     }
 }
@@ -559,6 +559,9 @@ async def handle_upright_challenge(data: dict) -> dict:
                 'processing_time': (datetime.now() - start_time).total_seconds()
             }
         
+        # Debug: Log model configuration
+        logger.info(f"🔍 UPRIGHT DEBUG: Model config: input_size={model_data['input_size']}, confidence={model_data['confidence']}, nms={model_data['nms']}")
+        
         image_bytes = base64.b64decode(image_b64.split(',')[1])
         reconstructed_image_pil = Image.open(io.BytesIO(image_bytes))
         original_w, original_h = reconstructed_image_pil.size
@@ -566,10 +569,11 @@ async def handle_upright_challenge(data: dict) -> dict:
         # Debug: Log image dimensions
         logger.info(f"🔍 UPRIGHT DEBUG: Original image dimensions: {original_w}x{original_h}")
 
-        # Determine input size for this inference (prefer model config, default to 300 for upright 3x2 grid)
-        input_size = model_data.get('input_size')
-        if not isinstance(input_size, int) or input_size <= 0:
-            input_size = 300
+        # Use the model's configured input size consistently
+        input_size = model_data['input_size']
+        
+        # Debug: Log model configuration
+        logger.info(f"🔍 UPRIGHT DEBUG: Model configured input size: {input_size}")
 
         input_tensor = preprocess_image(image_bytes, input_size)
         outputs = model_data['session'].run(None, {model_data['input_name']: input_tensor})[0]
@@ -604,13 +608,25 @@ async def handle_upright_challenge(data: dict) -> dict:
         
         # Debug: Log scaling parameters
         logger.info(f"🔍 UPRIGHT DEBUG: Scaling parameters: scale={scale:.4f}, pad_x={pad_x:.2f}, pad_y={pad_y:.2f}")
-        logger.info(f"🔍 UPRIGHT DEBUG: Model input size: {model_data['input_size']}")
+        logger.info(f"🔍 UPRIGHT DEBUG: Input size used: {input_size}")
         
         x_center_orig = (box_model[0] - pad_x) / scale
         y_center_orig = (box_model[1] - pad_y) / scale
         
-        # Debug: Log original space coordinates
-        logger.info(f"🔍 UPRIGHT DEBUG: Original space coordinates: x_center={x_center_orig:.2f}, y_center={y_center_orig:.2f}")
+        # Debug: Log original space coordinates with detailed calculation
+        logger.info(f"🔍 UPRIGHT DEBUG: Coordinate transformation:")
+        logger.info(f"🔍 UPRIGHT DEBUG:   Model coordinates: x={box_model[0]:.2f}, y={box_model[1]:.2f}")
+        logger.info(f"🔍 UPRIGHT DEBUG:   Subtract padding: x={box_model[0]:.2f}-{pad_x:.2f}={box_model[0]-pad_x:.2f}, y={box_model[1]:.2f}-{pad_y:.2f}={box_model[1]-pad_y:.2f}")
+        logger.info(f"🔍 UPRIGHT DEBUG:   Divide by scale: x={box_model[0]-pad_x:.2f}/{scale:.4f}={x_center_orig:.2f}, y={box_model[1]-pad_y:.2f}/{scale:.4f}={y_center_orig:.2f}")
+        logger.info(f"🔍 UPRIGHT DEBUG: Final original space coordinates: x_center={x_center_orig:.2f}, y_center={y_center_orig:.2f}")
+        
+        # Validate coordinates are within reasonable bounds
+        if x_center_orig < 0 or y_center_orig < 0 or x_center_orig > original_w or y_center_orig > original_h:
+            logger.warning(f"⚠️ UPRIGHT WARNING: Coordinates out of bounds: ({x_center_orig:.2f}, {y_center_orig:.2f}) for image {original_w}x{original_h}")
+            # Clamp to image bounds
+            x_center_orig = max(0, min(x_center_orig, original_w))
+            y_center_orig = max(0, min(y_center_orig, original_h))
+            logger.info(f"🔧 UPRIGHT FIX: Clamped coordinates to: ({x_center_orig:.2f}, {y_center_orig:.2f})")
         
         # Debug: Log grid calculation details
         grid_cols, grid_rows = 3, 2