Are there python scripts for detection (inference) and/or tracking?

The batch size is not mentioned.

Actually, in your code, even the first training loss is nan. How could you "loss.backward()"? Besides, the output of your net is also "nan".

This is really a great job. When will the CMT code be updated

Why am I getting nan values in val loss, what could be the issue?

Isn't The code"result = x + self.conv2(self.dwconv(self.conv1(x)))" in module class IRFFN(nn.Module) of cmt_module.py "result = self.conv2(self.dwconv(self.conv1(x))+self.conv1(x))" according to the paper "CMT: Convolutional Neural Networks Meet Vision Transformers"?

Hello, dear author

I set up a random seed to run the network you wrote, but the training results of the two runs are different. Is there any randomness in your weight initialization

Training the cmt-b model with cifar10 results in lower accuracy than the cmt-ti model.
Is this the right thing? Am I wrong?

epoch为1时loss就已经显示为nan，top1只有10%左右