mAP too low when training yolov4 tiny custom model

[andreluizeng]

Hi there, I have been trying to train my custom yolov4 tiny model but it seems like there are some issues when calculating the mAP, also the avg loss is too high. If I get the same config and image dataset and use the 795474f0ed022fd73af4e70cac79751f672d8af9 commit, it works perfectly. Is there any known issue ?

cheers, Andre