Update app.py

app.py

@@ -676,55 +676,52 @@ def convert_image_to_3d():
                 return
             
             try:
-                def estimate_depth():
-                 with torch.no_grad():
-                  # Make sure image is in RGB format for the models
-                  rgb_image = image
-                  if rgb_image.mode == 'RGBA':
-                   # Convert RGBA to RGB for model processing
-                   rgb_image = Image.new('RGB', image.size, (255, 255, 255))
-                   rgb_image.paste(image, mask=image.split()[3])  # Use alpha channel as mask
-         
-                   # DPT-Large
-                   dpt_result = dpt_model(rgb_image)
-                   dpt_depth = dpt_result["depth"]
-                   processing_jobs[job_id]['progress'] = 40
-        
-                    # Depth Anything (if loaded)
-                   if da_model and da_processor:
-                      inputs = da_processor(images=rgb_image, return_tensors="pt")  # Use RGB image here
-                       inputs = {k: v.to("cpu") for k, v in inputs.items()}
-                         outputs = da_model(**inputs)
-                         da_depth = outputs.predicted_depth.squeeze()
-                         da_depth = torch.nn.functional.interpolate(
-                           da_depth.unsqueeze(0).unsqueeze(0),
-                           size=(image.height, image.width),
-                           mode='bicubic',
-                            align_corners=False
-                                 ).squeeze()
-                                 processing_jobs[job_id]['progress'] = 50
+ def estimate_depth():
+    with torch.no_grad():
+        # Make sure image is in RGB format for the models
+        rgb_image = image
+        if rgb_image.mode == 'RGBA':
+            # Convert RGBA to RGB for model processing
+            rgb_image = Image.new('RGB', image.size, (255, 255, 255))
+            rgb_image.paste(image, mask=image.split()[3])  # Use alpha channel as mask
+        
+        # DPT-Large
+        dpt_result = dpt_model(rgb_image)
+        dpt_depth = dpt_result["depth"]
+        processing_jobs[job_id]['progress'] = 40
+        
+        # Depth Anything (if loaded)
+        if da_model and da_processor:
+            inputs = da_processor(images=rgb_image, return_tensors="pt")  # Use RGB image here
+            inputs = {k: v.to("cpu") for k, v in inputs.items()}
+            outputs = da_model(**inputs)
+            da_depth = outputs.predicted_depth.squeeze()
+            da_depth = torch.nn.functional.interpolate(
+                da_depth.unsqueeze(0).unsqueeze(0),
+                size=(image.height, image.width),
+                mode='bicubic',
+                align_corners=False
+            ).squeeze()
+            processing_jobs[job_id]['progress'] = 50
             
-                               # Improved fusion of depth maps
-                               fused_depth = fuse_depth_maps(dpt_depth, da_depth, detail_level)
-                            else:
-                            # Just use DPT with enhanced processing if Depth Anything is not available
-                             fused_depth = np.array(dpt_depth) if isinstance(dpt_depth, Image.Image) else dpt_depth
-                               if len(fused_depth.shape) > 2:
-                                    fused_depth = np.mean(fused_depth, axis=2)
-                                 # Apply more conservative normalization
-                                  p_low, p_high = np.percentile(fused_depth, [2, 98])
-                                 fused_depth = np.clip((fused_depth - p_low) / (p_high - p_low), 0, 1) if p_high > p_low else fused_depth
-                                   # Apply compression to limit extreme depths
-                                     fused_depth = np.power(fused_depth, 0.85)
-        
-                                   # Save depth map for debugging
-                                    depth_debug_path = os.path.join(output_dir, "depth_map.png")
-                                        cv2.imwrite(depth_debug_path, (fused_depth * 255).astype(np.uint8))
-        
-                                         return fused_depth
-
-                        
-                        
+            # Improved fusion of depth maps
+            fused_depth = fuse_depth_maps(dpt_depth, da_depth, detail_level)
+        else:
+            # Just use DPT with enhanced processing if Depth Anything is not available
+            fused_depth = np.array(dpt_depth) if isinstance(dpt_depth, Image.Image) else dpt_depth
+            if len(fused_depth.shape) > 2:
+                fused_depth = np.mean(fused_depth, axis=2)
+            # Apply more conservative normalization
+            p_low, p_high = np.percentile(fused_depth, [2, 98])
+            fused_depth = np.clip((fused_depth - p_low) / (p_high - p_low), 0, 1) if p_high > p_low else fused_depth
+            # Apply compression to limit extreme depths
+            fused_depth = np.power(fused_depth, 0.85)
+        
+        # Save depth map for debugging
+        depth_debug_path = os.path.join(output_dir, "depth_map.png")
+        cv2.imwrite(depth_debug_path, (fused_depth * 255).astype(np.uint8))
+        
+        return fused_depth
                 
                 fused_depth, error = process_with_timeout(estimate_depth, [], TIMEOUT_SECONDS)