Question about ODVGdataset

[swagger-coder]

I have a question ODVGdataset. I find the label map of "VG" mode is updated in every image without a global label map. And all the classes of the image may in range [0, len(uni_caption_list)]. https://github.com/longzw1997/Open-GroundingDino/blob/main/datasets/odvg.py line 105 label_map = {} for idx in range(len(uni_caption_list)): label_map[uni_caption_list[idx]] = idx classes = [label_map[cap] for cap in caption_list] caption = ' . '.join(uni_caption_list) + ' .'