You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for your interest in our work. I am sorry for the delay. I have uploaded a version here, but the code can be messy.
Please find the code in the zip file and install the package following the README.md.
The training scripts are provided in the file, such as train_digit5_ms_cutmix.sh for training on Digit-Five using FixMatch-CM. The backbone models for the FixMatch-CM can be found at dassl/modeling/backbone, such as resnet_mixstyle.py or cnn_digit5_m3sda_mixstyle.py, all with the suffix ‘_mixstyle.py’.
After the training with FixMatch-CM, you will obtain the first-step trained model, such as model.pth.
The second step BORT2 training can use the script 'train_digit5_mscm_dist_net_meta_train_retrain.sh' (similar names apply to PACS or DomainNet). Note that in the sh script, you will need to specify the checkpoint from the first-step training by MODEL.INIT_WEIGHTS model.pth. The config files and the trainer (like FixMatchMSCMDistNetMetaLearnRetrain) are specified in the script. And the trainer for BORT2 is named FixMatchMSCMDistNetMetaLearnRetrain of which the py file can be found at dassl/engine/da/fixmatch_mscm_dist_net_meta_learn_retrain.py.
It says that this repo contains the PyTorch implementation of the paper but there are no files here.
The text was updated successfully, but these errors were encountered: