update

demo.py

@@ -265,14 +265,14 @@ def visual_grounding(image, text):
         attn_masks = []
         while i < max_len and unfinish_flag.any():
             # print(i)
-            prev_output_tokens_11_tensor = torch.tensor(np.array(prev_output_token_11)).to(img.device).long()
-            prev_output_tokens_12_tensor = torch.tensor(np.array(prev_output_token_12)).to(img.device).long()
-            prev_output_tokens_21_tensor = torch.tensor(np.array(prev_output_token_21)).to(img.device).long()
-            prev_output_tokens_22_tensor = torch.tensor(np.array(prev_output_token_22)).to(img.device).long()
-            delta_x1_tensor = torch.tensor(np.array(delta_x1)).to(img.device)
-            delta_x2_tensor = torch.tensor(np.array(delta_x2)).to(img.device)
-            delta_y1_tensor = torch.tensor(np.array(delta_y1)).to(img.device)
-            delta_y2_tensor = torch.tensor(np.array(delta_y2)).to(img.device)
+            prev_output_tokens_11_tensor = torch.tensor(np.array(prev_output_token_11)).long()
+            prev_output_tokens_12_tensor = torch.tensor(np.array(prev_output_token_12)).long()
+            prev_output_tokens_21_tensor = torch.tensor(np.array(prev_output_token_21)).long()
+            prev_output_tokens_22_tensor = torch.tensor(np.array(prev_output_token_22)).long()
+            delta_x1_tensor = torch.tensor(np.array(delta_x1)).float()
+            delta_x2_tensor = torch.tensor(np.array(delta_x2)).float()
+            delta_y1_tensor = torch.tensor(np.array(delta_y1)).float()
+            delta_y2_tensor = torch.tensor(np.array(delta_y2)).float()
 
             net_output = model.decoder(
                 prev_output_tokens_11_tensor,