search

kiwi12138 / RealisticTTA

Official repository for AAAI2024 paper <Unraveling Batch Normalization for Realistic Test-Time Adaptation>.
6 stars 1 forks source link

α-BN Results #1

Open tuwangwenx opened 4 months ago

tuwangwenx commented 4 months ago

In Table 1, the batch size is 4, 2, 1. The results I ran are completely different from yours. Could you give the forward_and_adapt function you have here. I directly used the code given by Online Test-time Adaptation. I also modified self.norm = nn.BatchNorm2d(self.layer.num_features, affine=False, momentum=cfg.TEST.MOMENTUM, track_running_stats=True) in AlphaBatchNorm. image

kiwi12138 commented 4 months ago

If you would like to reproduce the Alpha_BN results, you can directly follow the original code of Online Test-time Adaptationhttps://github.com/mariodoebler/test-time-adaptation/tree/main. Just modify the batchsize parameter in norm_alpha.yaml and run. You don't need to modify any other functions.

tuwangwenx commented 4 months ago

If you would like to reproduce the Alpha_BN results, you can directly follow the original code of Online Test-time Adaptationhttps://github.com/mariodoebler/test-time-adaptation/tree/main. Just modify the batchsize parameter in norm_alpha.yaml and run. You don't need to modify any other functions.

Thank you for your reply and your paper. Could you please provide me with the AdaptBN code?