PhoenixZ810/MG-LLaVA

duplicate sample 'id'

VeritasXu opened this issue · 2 comments

Dear authors,

Many thanks for sharing the interesting work. However, there seem to be many samples with the same 'id' (about 250k) in bbox.json and annotation.json (for train) , resulting in mismatched bboxes and images.

Thank you for bringing this to our attention!
Upon re-examining the fine-tuning annotation JSON, I have confirmed that there are indeed duplicate IDs present. Nevertheless, it appears that the same ID corresponds to different conversations yet identical images, indicating that the bounding boxes and images remain properly aligned.

Got that, thank you.