search

lyuwenyu / RT-DETR

[CVPR 2024] Official RT-DETR (RTDETR paddle pytorch), Real-Time DEtection TRansformer, DETRs Beat YOLOs on Real-time Object Detection. 🔥 🔥 🔥
Apache License 2.0
2.61k stars 303 forks source link

Clarification about label preprocessing #491

Closed amirbakhtiary23 closed 9 hours ago

amirbakhtiary23 commented 2 days ago

Hi, thank you for your great contribution. As you have mentioned in #99 target boxes are converted to XYXY then to CXCY. But i cannot find exactly where does this conversion to CXCY happens. Can you please point it out to me? I found where the CXCY to XYXY take place for computing loss value, but I need to write my own Dataloader in order to train on my custom fusion dataset so I need to know where does the first conversion take place.

lyuwenyu commented 11 hours ago
  1. config: https://github.com/lyuwenyu/RT-DETR/blob/main/rtdetrv2_pytorch/configs/rtdetrv2/include/dataloader.yml#L14
  2. code: https://github.com/lyuwenyu/RT-DETR/blob/main/rtdetrv2_pytorch/src/data/transforms/_transforms.py#L93-L113
amirbakhtiary23 commented 9 hours ago

Thank you very much.