why the test result and valid result on the same dataset is different?

[Carson-T]

When runing train.py, we got the valid result on valid set. Then I use the best.pth to run test.py on the same valid result, but the result is different from the one during training. The mAP is much lower. What is the problem? I use hyp.scratch.p5.yaml as the training parameter.

[Zeev1988]

I have the same issue. Doesn't the training script uses best.pt weights?