Something wrong when evaluate

[jy0x4f]

rtdetr_pytorch/src/solver/det_engine.py in line 136， When my training data is 1280x720 and the data is preprocessed to 640x640, I found that when calculating the metric, this code returns the predicted bbox from the original data size (1280x720), but the bbox in the label corresponds to the preprocessed 640x640 sized value, which is then sent directly to the calculation of the mAP, where do I process out the Problems?

[lyuwenyu]

This target has not been used to compute metric, just for extracting image_id info.

---