Refactor image processing code to use torch.no_grad() context manager

app.py

@@ -93,22 +93,23 @@ with gr.Blocks(css=css) as demo:
         image = Normalize(mean=[0.485, 0.456, 0.406], std=[
             0.229, 0.224, 0.225])(image)
         # image = torch.from_numpy(image).unsqueeze(0)
-        image = torch.autograd.Variable(image.unsqueeze(0))
+        with torch.no_grad():
+            image = torch.autograd.Variable(image.unsqueeze(0))
 
-        pred_depths_r_list, _, _ = model(image)
-        image_flipped = flip_lr(image)
-        pred_depths_r_list_flipped, _, _ = model(image_flipped)
-        pred_depth = post_process_depth(
-            pred_depths_r_list[-1], pred_depths_r_list_flipped[-1])
+            pred_depths_r_list, _, _ = model(image)
+            image_flipped = flip_lr(image)
+            pred_depths_r_list_flipped, _, _ = model(image_flipped)
+            pred_depth = post_process_depth(
+                pred_depths_r_list[-1], pred_depths_r_list_flipped[-1])
 
-        pred_depth = pred_depth.cpu().numpy().squeeze()
-        colored_depth = cv2.applyColorMap(
-            pred_depth, cv2.COLORMAP_INFERNO)[:, :, ::-1]
+            pred_depth = pred_depth.cpu().numpy().squeeze()
+            colored_depth = cv2.applyColorMap(
+                pred_depth, cv2.COLORMAP_INFERNO)[:, :, ::-1]
 
-        tmp = tempfile.NamedTemporaryFile(suffix='.png', delete=False)
-        pred_depth.save(tmp.name)
+            tmp = tempfile.NamedTemporaryFile(suffix='.png', delete=False)
+            pred_depth.save(tmp.name)
 
-        return [(original_image, colored_depth), tmp.name]
+            return [(original_image, colored_depth), tmp.name]
 
     submit.click(on_submit, inputs=[input_image], outputs=[
                  depth_image_slider, raw_file])