update app.py

app.py

@@ -64,18 +64,28 @@ def get_outputs(
 ):
     #  and torch.cuda.amp.autocast(dtype=torch.float16)
     with torch.inference_mode():
-        outputs = model.generate(
-            batch_images,
-            input_ids,
+        outputs = model(
+            vision_x=batch_images,
+            lang_x=input_ids,
             attention_mask=attention_mask,
-            max_new_tokens=max_generation_length,
-            min_length=min_generation_length,
-            num_beams=num_beams,
-            length_penalty=length_penalty,
-            image_start_index_list=image_start_index_list,
+            labels=None,
             image_nums=image_nums,
-            bad_words_ids=bad_words_ids,
+            image_start_index_list=image_start_index_list,
+            added_bbox_list=None,
+            add_box=False,
         )
+        # outputs = model.generate(
+        #     batch_images,
+        #     input_ids,
+        #     attention_mask=attention_mask,
+        #     max_new_tokens=max_generation_length,
+        #     min_length=min_generation_length,
+        #     num_beams=num_beams,
+        #     length_penalty=length_penalty,
+        #     image_start_index_list=image_start_index_list,
+        #     image_nums=image_nums,
+        #     bad_words_ids=bad_words_ids,
+        # )
 
     return outputs
 
@@ -145,13 +155,11 @@ def generate(
     scores = outputs["scores"]
     if len(scores) > 0:
         box = boxes[scores.argmax()]
-        iou = get_iou(box, gt_box)
     else:
         iou = 0.0
         # tqdm.write(f"output: {tokenizer.batch_decode(outputs)}")
         tqdm.write(f"no output for: {uniq_id}, {image_id}, {text}")
-    if iou >= 0.5:
-        correct += 1
+    print(f"{box}")
 
 
     gen_text = tokenizer.batch_decode(outputs)