Fix person classification model inference

- Fixed input dimensions to 448x640 (was incorrectly 640x480)
- Added better error handling with full traceback
- Fixed quantization handling to check if quantization info exists
- Added debug output for model input/output shapes and dtypes
- Updated UI to show correct input dimensions

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

app.py

@@ -32,6 +32,9 @@ output_details = interpreter.get_output_details()
 class_labels = ["No Person", "Person"]
 
 print(f"Model loaded successfully! Input shape: {input_details[0]['shape']}")
+print(f"Input dtype: {input_details[0]['dtype']}")
+print(f"Output shape: {output_details[0]['shape']}")
+print(f"Output dtype: {output_details[0]['dtype']}")
 print(f"Number of classes: {len(class_labels)}")
 print(f"SRAM-optimized model also available: {SRAM_MODEL_FILE}")
 # Force rebuild with modern design
@@ -41,8 +44,8 @@ def preprocess_image(image):
     """
     Preprocess image for Person Classification INT8 quantized model.
     """
-    # Resize to 640x480 as expected by the model (input is 448x640 but processes 640x480)
-    image = image.resize((640, 480))
+    # Resize to 448x640 as expected by the model
+    image = image.resize((448, 640))
     
     # Convert to numpy array
     img_array = np.array(image, dtype=np.float32)
@@ -87,10 +90,14 @@ def classify_image(image):
         predictions = output_data[0]  # Remove batch dimension
         
         # Convert from INT8 quantized output to probabilities
-        # Dequantize the output
-        scale = output_details[0]['quantization'][0]
-        zero_point = output_details[0]['quantization'][1]
-        predictions = scale * (predictions.astype(np.float32) - zero_point)
+        # Dequantize the output if quantization info is available
+        if 'quantization' in output_details[0] and output_details[0]['quantization'] is not None:
+            scale = output_details[0]['quantization'][0]
+            zero_point = output_details[0]['quantization'][1]
+            predictions = scale * (predictions.astype(np.float32) - zero_point)
+        else:
+            # If no quantization info, assume output is already in correct format
+            predictions = predictions.astype(np.float32)
         
         # For binary classification, get the probability
         # The model outputs a single value for person probability
@@ -112,8 +119,11 @@ def classify_image(image):
         
         return result_text
         
-    except Exception:
-        return "Error processing image. Please try again."
+    except Exception as e:
+        import traceback
+        error_msg = f"Error processing image: {str(e)}\n\nTraceback:\n{traceback.format_exc()}"
+        print(error_msg)  # Log to console
+        return f"Error processing image: {str(e)}"
 
 def load_example_image(example_path):
     """Load example images for demonstration."""
@@ -368,7 +378,7 @@ with gr.Blocks(
                                 SRAM: ~1.5 MB<br>
                                 Flash: ~1.5 MB<br>
                                 Model: Person Classifier<br>
-                                Input: 640×480×3
+                                Input: 448×640×3
                             </div>
                         </div>
                     </div>