The impact of the order of elements (i.e., text) within "cat_list" used in model evaluation about open-groundingdino HOT 1 CLOSED

1. Training and testing need to have the same cat_list. It’s easy to understand if you treat it as traditional Object Detection task.
2. The order of text elements in cat_list may not be crucial, but be consistent, just like traditional Object Detection, it depends on your anotations.
3. Fine-tuning on OD datasets is not the main purpose, this will reduce the generalization ability of the model, and additional strategies need to be used to maintain generalization. #26
4. I don't think fine-tuning GDINO model will get the best performance(mAP, ACC, etc.) if you only want to detect object in a closed set(not a open-set), perhaps you can try some other networks(co-detr, EVA, DINO,internimage, etc. )

from open-groundingdino.