koncle/MVDG

Introduction

The implementation of ECCV 2022 paper MVDG: A Unified Multi-view Framework for Domain Generalization.

Install packages

pytorch version >= 1.8， otherwise, some errors may occur.

conda install pytorch torchvision cudatoolkit
conda install matplotlib tqdm tensorboardX

Run the code

The core part of our method can be found in

There are two implementations of MVRML: a vanilla version and a paralleled version. The paralleled version could train faster but requires more GPUs (same as the number trajectories).

To run MVRML, you could run the following code. The --data-root argument needs to be changed according to your folder.

bash script/mvrml.sh

After the model has been trained, MVP could be applied to the trained model with the following code:

bash script/MVP.sh