Awesome

pytorch-domain-adaptation

This is an unofficial pytorch implementation of algorithms for domain adaptation.

Note that this is an ongoing project and I cannot fully reproduce the results. Suggestions are welcome!

List of algorithms

• From source to target and back: symmetric bi-directional adaptive GAN [Russo+, CVPR2018].
• Augmented Cyclic Adversarial Learning for Domain Adaptation [Hosseini-Asl+, arXiv2018].

Requirements

• Python 3.5+
• PyTorch 0.4
• TorchVision
• TensorboardX
• batchup
• click

Usage

These examples are for the MNIST to USPS experiment.

Train Source Only Model

CUDA_VISIBLE_DEVICES=<gpu_id> python train_classifier.py --exp mnist_usps --train_type unsup

Train Target Only Model

CUDA_VISIBLE_DEVICES=<gpu_id> python train_classifier.py --exp mnist_usps --train_type sup

Train Model

UDA_VISIBLE_DEVICES=<gpu_id> python test_classifier.py --exp mnist_usps --snapshot <snapshot_dir>