Verifying that update_param in the paper=='all'

SakurajimaMaiii / TSD

[CVPR 2023] Feature Alignment and Uniformity for Test Time Adaptation

https://arxiv.org/abs/2303.10902

MIT License

41 stars 1 forks source link

Verifying that update_param in the paper=='all' #8

Closed YunYunY closed 1 year ago

YunYunY commented 1 year ago

Hi, I want to verify with the author that the model performance reported in the paper is always by updating the whole backbone parameters right? Otherwise, line 198 in unsupervise_adapt.py looks like a tiny bug to me.

SakurajimaMaiii commented 1 year ago

We update all trainable parameters as we claimed in paper Sec 4.1 implementation. Unlike [45, 66], all the trainable layers are updated and no special selection is required in our method.