Huawei-competition-feature-separation-in-cross-domain

Huawei competition: feature separation in cross domain

1. Basic operating environment: Python>=3.6.7, tensorflow=1.15
2. Although the running framework of the model is tensorflow, the framework of cmnist data generation module (gen_color_mnist.py) is pytorch. Please make sure you install it.

Preparation works: Run gen_color_mnist.py to generate cmnist, and then you can find train_x.npy, train_y.npy, test_x.npy and test_y.npy in the path data_sets/cmnist/fgbg_cmnist_cpr/. Put the generated four files in the path data/.

Training:

1. If you want to extract stable features in CMNIST data, please run train_stable.py. If you fall into the local optimum during training, please retrain.
2. If you want to extract unstable features on the CMNIST test set, please run train_unstable.py. Before running, please make sure that the model file generated by train_stable.py had been sved in model/.