ispc-lab / bmd Goto Github PK

[ECCV 2022] The official repository of our paper "BMD: A General Class-balanced Multicentric Dynamic Prototype Strategy for Source-free Domain Adaptation"

Python 98.66% Shell 1.34%

unsupervised-domain-adaptation source-free-domain-adaptation

bmd's People

Contributors

Stargazers

Watchers

bmd's Issues

Can not reproduce office-31 and officehome dataset results

Hi, thanks for your excellent work.
I can't reproduce the results based on the office-31 and officehome dataset. I have changed the parameter S, α=0.3，β=0.1 according to the article. Are there another parameters I need to change? How can I solve that?

Office31 results

Good morning, I am trying to reproduce the Office31 results of the paper but I am getting lower accuracy for every domain-pair. I am using the released code and the hparams written in the paper. Do you have any suggestions on what can be the cause of this? What version of CUDA did you use?
Thank you

Most parameters are frozen

BMD/main_target.py

Lines 184 to 188 in 3723b32

 for k, v in model.backbone_layer.named_parameters(): 

 if "bn" in k: 

 param_group += [{'params': v, 'lr': args.lr*0.1}] 

 else: 

 v.requires_grad = False

Hello, thanks for sharing your code. In the above a few lines, only parameters in BN layers will be updated and all the other parameters are frozen. Does this mean only BN parameters are trained?

Recommend Projects

ispc-lab / bmd Goto Github PK

bmd's People

Contributors

Stargazers

Watchers

bmd's Issues

Can not reproduce office-31 and officehome dataset results

Office31 results

Most parameters are frozen

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	for k, v in model.backbone_layer.named_parameters():
	if "bn" in k:
	param_group += [{'params': v, 'lr': args.lr*0.1}]
	else:
	v.requires_grad = False