search

lhoyer / DAFormer

[CVPR22] Official Implementation of DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation
Other
470 stars 93 forks source link

Request for pretrained model of Table 1 #12

Closed super233 closed 2 years ago

super233 commented 2 years ago

Hi, thanks for your awesome work. Can you provide the pretrained model of SegFormer for UDA (58.2) in Table 1?

image

super233 commented 2 years ago

One more question, I have noticed that your pretrained models are about BN but not SyncBN, however, the official pretrained models are all about SyncBN.

Have you use some ways to convert model weights from official pretrained models about SyncBn into BN? Or have you trained the model with BN from scratch?

I'm looking forward to your reply. :-)

lhoyer commented 2 years ago

The models use BN instead of SyncBN as only one GPU is used for training. We have trained the models of Tab. 1 ourselves. You can replicate the training using python run_experimenty.py --exp 1. Please have a look at the experiments.py#L226 for more details.

lhoyer commented 2 years ago

I have uploaded the requested checkpoint to Google Drive: https://drive.google.com/file/d/1neppvQVH2kqjS2KSS82llk9iZ3QNry4h/view?usp=sharing

super233 commented 2 years ago

Thanks very much!