Update app.py

app.py

@@ -210,42 +210,43 @@ def count_objects(image, pos_caption, neg_caption, box_threshold, point_radius,
     use_neg = bool(neg_caption and neg_caption.strip() and neg_caption != '.')
     
 
-    if use_neg:
+    if not use_neg:
         # print('neg_caption: ', neg_caption)
-        neg_inputs = processor(
-            images=image,
-            text=neg_caption,
-            return_tensors="pt",
-            padding=True
-        )
-        neg_inputs = {k: v.to(device) for k, v in neg_inputs.items()}
-        neg_inputs['pixel_values'] = neg_inputs['pixel_values'].to(torch.bfloat16)
-
-        # Add negative inputs to positive inputs dict
-        pos_inputs['neg_token_type_ids'] = neg_inputs['token_type_ids']
-        pos_inputs['neg_attention_mask'] = neg_inputs['attention_mask']
-        pos_inputs['neg_pixel_mask'] = neg_inputs['pixel_mask']
-        pos_inputs['neg_pixel_values'] = neg_inputs['pixel_values']
-        pos_inputs['neg_input_ids'] = neg_inputs['input_ids']
-        pos_inputs['use_neg'] = True
-    else:
         neg_caption = "None."
-        neg_inputs = processor(
-            images=image,
-            text=neg_caption,
-            return_tensors="pt",
-            padding=True
-        )
-        neg_inputs = {k: v.to(device) for k, v in neg_inputs.items()}
-        neg_inputs['pixel_values'] = neg_inputs['pixel_values'].to(torch.bfloat16)
-
-        # Add negative inputs to positive inputs dict
-        pos_inputs['neg_token_type_ids'] = neg_inputs['token_type_ids']
-        pos_inputs['neg_attention_mask'] = neg_inputs['attention_mask']
-        pos_inputs['neg_pixel_mask'] = neg_inputs['pixel_mask']
-        pos_inputs['neg_pixel_values'] = neg_inputs['pixel_values']
-        pos_inputs['neg_input_ids'] = neg_inputs['input_ids']
-        pos_inputs['use_neg'] = False
+    neg_inputs = processor(
+        images=image,
+        text=neg_caption,
+        return_tensors="pt",
+        padding=True
+    )
+    neg_inputs = {k: v.to(device) for k, v in neg_inputs.items()}
+    neg_inputs['pixel_values'] = neg_inputs['pixel_values'].to(torch.bfloat16)
+
+    # Add negative inputs to positive inputs dict
+    pos_inputs['neg_token_type_ids'] = neg_inputs['token_type_ids']
+    pos_inputs['neg_attention_mask'] = neg_inputs['attention_mask']
+    pos_inputs['neg_pixel_mask'] = neg_inputs['pixel_mask']
+    pos_inputs['neg_pixel_values'] = neg_inputs['pixel_values']
+    pos_inputs['neg_input_ids'] = neg_inputs['input_ids']
+    pos_inputs['use_neg'] = True
+    # else:
+    #     neg_caption = "None."
+    #     neg_inputs = processor(
+    #         images=image,
+    #         text=neg_caption,
+    #         return_tensors="pt",
+    #         padding=True
+    #     )
+    #     neg_inputs = {k: v.to(device) for k, v in neg_inputs.items()}
+    #     neg_inputs['pixel_values'] = neg_inputs['pixel_values'].to(torch.bfloat16)
+
+    #     # Add negative inputs to positive inputs dict
+    #     pos_inputs['neg_token_type_ids'] = neg_inputs['token_type_ids']
+    #     pos_inputs['neg_attention_mask'] = neg_inputs['attention_mask']
+    #     pos_inputs['neg_pixel_mask'] = neg_inputs['pixel_mask']
+    #     pos_inputs['neg_pixel_values'] = neg_inputs['pixel_values']
+    #     pos_inputs['neg_input_ids'] = neg_inputs['input_ids']
+    #     pos_inputs['use_neg'] = False
 
     # Run inference
     with torch.no_grad():