test input

app.py

@@ -103,13 +103,26 @@ def deepseek(image, text_input):
     )
 
     answer = tokenizer.decode(outputs[0].cpu().tolist(), skip_special_tokens=False)
+
+    object_ref_pattern = r"<\|object_ref_start\|>(.*?)<\|object_ref_end\|>"
+    det_pattern = r"<\|det\|>\[\[(.+)]]<\|\/det\|>"
+
+    det_content = re.search(det_pattern, answer).group(1)
+    bbox = [ int(v.trim()) / 999 for v in det_content.split(",")]
+    w, h = image.size()
+    bbox = [bbox[0] * w, bbox[1] * h, bbox[2] * w, bbox[3] * h] 
+    answer
     print(f"{prepare_inputs['sft_format'][0]}", answer)
 
+    
+    scaled_boxes = rescale_bounding_boxes(bbox, image.width, image.height)
+    return text_input, scaled_boxes, draw_bounding_boxes(image, scaled_boxes)
+
 
 @spaces.GPU
 def run_example(image, text_input, model_id="OS-Copilot/OS-Atlas-Base-7B"):
 
-    deepseek(image, text_input)
+    return deepseek(image, text_input)
     
 
 def run_example_old(image, text_input, model_id="OS-Copilot/OS-Atlas-Base-7B"):