A PyTorch implementation of 'Learning Transferable Features with Deep Adaptation Networks'. The contributions of this paper are summarized as fol- lows.

• They propose a novel deep neural network architecture for domain adaptation, in which all the layers corresponding to task-specific features are adapted in a layerwise manner, hence benefiting from “deep adaptation.”
• They explore multiple kernels for adapting deep representations, which substantially enhances adaptation effectiveness compared to single kernel methods. Our model can yield unbiased deep features with statistical guarantees.

• python 3
• pytorch 0.3.1
• torchvision 0.2.0

1. You can download Office31 dataset here. And then unrar dataset in ./dataset/.
2. You can change the source_name and target_name in DAN.py to set different transfer tasks.
3. Run python DAN.py.