Python 3.6.9, Pytorch 1.6.0, Torch Vision 0.7.0. To track the training details, we also used neptune, but this is optional configuration. You also need sklearn (0.23.2 is used).
For dataset preparation, please follow nc_ps/README.md
We are using the same datasets. Do not forget to put a dataset and txt file link in this directory (./data, ./txt).
Scripts are stored in scripts directory. Please change the random seed when testing on different source validation samples.
MCC
sh scripts/train_a2d_mcc.sh
CDAN
sh scripts/train_a2d.sh
Note that only in the experiments of MCC on visda, we employ ResNet101 following their paper.
The large proportion of this directory is borrowed from CDAN and MCC.