Tried using some modules from sam6d but got poor results after setting model.eval() during testing

[W-QY]

First of all thank you very much for your work!

I tried to add the "Fine Point Matching" module in your code to my network. During the training process, the loss converged very well, but the test results were poor. After I removed "model.eval()", the model test results were very good. I would like to ask you if you know the reason?

[huassaue]

是不是batch_size设置过小