Hi, Thanks for this great work and for sharing the code.

I am currently trying to reproduce the result for the tennis dataset using exactly the same code and configuration provided.

However, I am experiencing a performance drop where for the RegNety_200 model, the mAP dropped to 85 for the test dataset. When I trained the model myself. When I load the checkpoint, the performance also dropped to 90 instead of 96 where I did not change anything.

Just wondering, are there small details that I missed out on that could have triggered this problem? I am using a different machine A100, but I did not expect such a performance drop.

Please help me!

Thank you.