Update utils.py

utils.py

@@ -160,13 +160,13 @@ def generate_similiarity_map(images, attn_map, all_bpe_strings, vis_list, target
     attn_norm = get_similarity_map(attn_map.unsqueeze(0), (target_height, target_width), min_max=True, threshold=0.15)
     print("attn_norm ",attn_norm.shape) # 有问题attn_norm  torch.Size([1, 4, 448, 448])
     print('all_bpe_strings:{:}'.format(all_bpe_strings))
-    indexes_without_space = torch.tensor([index for index, string in enumerate(all_bpe_strings) if ' ' is not string])
+    # indexes_without_space = torch.tensor([index for index, string in enumerate(all_bpe_strings) if ' ' is not string])
 
     # Draw similarity map
     # print(images_vis.shape)
     images_vis = (images_vis.permute(1,2,0).cpu().numpy() * 125).astype('uint8')
     for b in range(attn_norm.shape[0]):
-        for n in range(attn_norm.shape[1]-1):
+        for n in range(attn_norm.shape[1]):
             vis = (attn_norm[b, n, :, :].float().detach().cpu().numpy() * 255).astype('uint8')
             vis = cv2.applyColorMap(vis, cv2.COLORMAP_JET)
             print("images_vis",images_vis.shape)
@@ -176,17 +176,17 @@ def generate_similiarity_map(images, attn_map, all_bpe_strings, vis_list, target
             vis = cv2.resize(vis, src_iamge_size)
             vis_list.append(vis)  # Add each visualization to the list
         
-        without_space_norm = attn_norm[b, indexes_without_space, :, :].max(0)[0]
-        space_norm = attn_norm[b, -1, :, :]
-        all_attn_norm = without_space_norm - space_norm
-        print(f'min:{all_attn_norm.min()};max:{all_attn_norm.max()}')
-        all_attn_norm = (all_attn_norm - all_attn_norm.min()) / (all_attn_norm.max() - all_attn_norm.min())
-        all_attn_norm = (all_attn_norm.float().detach().cpu().numpy() * 255).astype('uint8')
-        vis = cv2.applyColorMap(all_attn_norm, cv2.COLORMAP_JET)
-        vis = images_vis * 0.5 + vis * 0.5
-        vis = cv2.cvtColor(vis.astype('uint8'), cv2.COLOR_BGR2RGB)
-        vis = cv2.resize(vis, src_iamge_size)
-        vis_list.append(vis)  # Add each visualization to the list
+        # without_space_norm = attn_norm[b, indexes_without_space, :, :].max(0)[0]
+        # space_norm = attn_norm[b, -1, :, :]
+        # all_attn_norm = without_space_norm - space_norm
+        # print(f'min:{all_attn_norm.min()};max:{all_attn_norm.max()}')
+        # all_attn_norm = (all_attn_norm - all_attn_norm.min()) / (all_attn_norm.max() - all_attn_norm.min())
+        # all_attn_norm = (all_attn_norm.float().detach().cpu().numpy() * 255).astype('uint8')
+        # vis = cv2.applyColorMap(all_attn_norm, cv2.COLORMAP_JET)
+        # vis = images_vis * 0.5 + vis * 0.5
+        # vis = cv2.cvtColor(vis.astype('uint8'), cv2.COLOR_BGR2RGB)
+        # vis = cv2.resize(vis, src_iamge_size)
+        # vis_list.append(vis)  # Add each visualization to the list
 
     return vis_list