Run python fd_classifier_cmnist.py.

We provide a pretrained model here: cvpr_cmnist_s8_z32_fdc_l_256_model_best_508.pth

Download waterbird dataset, we provide a version here

If you want to run our pretrained model, here are two models by running two times: Model 1, Models 2

Run python fd_waterbird.py for our experiment, you can choose to retrain your own or evaluating our downloaded checkpoints in Line 253 and L467. Also, change the data path to your saved data directory.

For SimCLR model, first download the pretrained model here, download ImageNet Rendition and ImageNet Sketch, then

Run CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python causal_imagenet_SSL.py --drop_xp --lr-max 3e-4.

For baseline:

Run CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python baseline_imagenet_SSL.py.