what's the difference between mm-grounding dino and the open-grounding dino? i found the odvg.py some difference

[lyf6]

in the flickr30ke2odvg.py of open-grounding dino, each region have a bbox and phrase, so why odvg.py in mm-grounding dino each region have multiple bbox? for i, region in enumerate(regions): bbox = region['bbox'] phrase = region['phrase'] tokens_positive = region['tokens_positive'] if not isinstance(bbox[0], list): bbox = [bbox] for box in bbox: instance = {} x1, y1, x2, y2 = box inter_w = max(0, min(x2, data['width']) - max(x1, 0)) inter_h = max(0, min(y2, data['height']) - max(y1, 0)) if inter_w * inter_h == 0: continue if (x2 - x1) < 1 or (y2 - y1) < 1: continue instance['ignore_flag'] = 0 instance['bbox'] = box instance['bbox_label'] = i phrases[i] = { 'phrase': phrase, 'tokens_positive': tokens_positive }

[lyf6]

i see, this code have modified the ccovert formate code