Re-Parameterizing for switching

DingXiaoH / RepLKNet-pytorch

Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs (CVPR 2022)

MIT License

860 stars 86 forks source link

Re-Parameterizing for switching #12

Closed BoltenWang-Meta closed 2 years ago

BoltenWang-Meta commented 2 years ago

How can i turn off Rep in training and turn it on in validation?

DingXiaoH commented 2 years ago

Hi, I don't think that makes sense, because the parameters of the small kernels need to be trained, and the purpose of Rep is to simplify the model and speed up the inference.