Update app.py

app.py

@@ -41,7 +41,6 @@
 #     demo.launch()
 
 
-
 import gradio as gr
 import cv2
 import numpy as np
@@ -50,27 +49,24 @@ import time
 import gc
 from inspector_engine import AdvancedBlockInspector
 
-# 1. Initialize engine with lazy loading enabled
-# On HF Spaces, we want to avoid loading everything into RAM until the first request
-inspector = AdvancedBlockInspector(yolo_model_path='yolo26n-obb.onnx')
+# Initialize engine with lazy loading
+inspector = AdvancedBlockInspector(yolo_model_path='yolo26n-obb.pt')
 
 def inspect(image):
+    """Main inspection function"""
     if image is None:
         return None, {"error": "No image uploaded"}
     
     try:
         start_time = time.time()
         
-        # 2. Convert Gradio (RGB) to OpenCV (BGR)
+        # Convert Gradio (RGB) to OpenCV (BGR)
         frame = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
         
-        # 3. Process image 
-        # The engine handles internal lazy loading of YOLO/Torch on first call
+        # Process image
         result = inspector.inspect_block(frame)
         
-        # 4. Correct Visualization Call
-        # Your engine requires (image, saddle_rois, result_details)
-        # result.saddle_results contains the status for each saddle
+        # Visualization
         vis_frame = frame.copy()
         if hasattr(inspector, 'last_saddles') and result.saddle_results:
             vis_frame = inspector.visualize_results(
@@ -79,26 +75,26 @@ def inspect(image):
                 result.saddle_results
             )
         
-        # 5. Convert back to RGB for Gradio
+        # Convert back to RGB for Gradio
         vis_rgb = cv2.cvtColor(vis_frame, cv2.COLOR_BGR2RGB)
         
-        # 6. Prepare JSON data
+        # Prepare JSON data
         res_dict = result.to_dict()
         res_dict['server_side_time_ms'] = (time.time() - start_time) * 1000
         
-        # 7. Explicit Memory Cleanup
-        # Helps keep the Space running on the free 16GB tier without OOM
-        del frame
+        # Memory cleanup
+        del frame, vis_frame
         gc.collect()
         
         return vis_rgb, res_dict
-
+        
     except Exception as e:
-        print(f"Error during inspection: {str(e)}")
+        import traceback
+        error_msg = f"Error: {str(e)}\n{traceback.format_exc()}"
+        print(error_msg)
         return None, {"error": str(e)}
 
-# 8. Define the Interface with API Name
-# Adding 'api_name' allows your Render App to call /predict directly
+# Create Gradio Interface
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# TMTL Industrial Inspector")
     gr.Markdown("Remote AI Inference Engine for Saddle Defect Detection")
@@ -111,18 +107,26 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         with gr.Column():
             output_img = gr.Image(type="numpy", label="AI Visualization")
             output_json = gr.JSON(label="Detailed Analysis")
-
+    
+    # Wire up the button - FIXED: removed api_name from click event
     btn.click(
         fn=inspect, 
         inputs=input_img, 
-        outputs=[output_img, output_json],
-        api_name="predict"  # CRITICAL: This allows Render to connect
+        outputs=[output_img, output_json]
     )
+    
+    # Add API endpoint properly - CRITICAL FIX
+    demo.load(api_name=False)  # Disable auto API name on load
 
 if __name__ == "__main__":
-    # 9. Performance tuning for Spaces
-    # queue() is essential for handling multiple users/Render requests concurrently
-    demo.queue(max_size=10).launch(
-        server_name="0.0.0.0", 
-        show_error=True
+    demo.queue(
+        max_size=10,
+        default_concurrency_limit=4  # Limit concurrent requests
+    ).launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        show_error=True,
+        share=False,
+        # CRITICAL: Enable API mode for external calls
+        show_api=True
     )