This is the PyTorch code for validating our proposed method in this paper. For simple validation purpose, only a trained model and the validation code are provided for review, the full code will be published later. Here we take the task A->D on Office-31 dataset as an example.
The framework of the proposed method is shown in Fig.2 of the manuscript.
- python == 3.6.10
- pytorch == 1.5.0
- cuda == 9.2
- torchvision = 0.6.0
- "network.py": contains the network architectures we used in this paper (including the backbone, the bottleneck, the classifier and the perturbation generator).
- "data": contains the list files and the images of the testing data.
- "data_list.py" re-implements the ImageList class in pytorch for loading the data with corresponding index.
- "loss.py": contains some loss functions we used in this paper.
- "AD": contains the trained models (F, B and C) of the task A->D for validation.
Limited by volume, we do not provide the Office-31 dataset. To validate the model, you may get the dataset here and put it to the path according to "dslr_list.txt". After that, just run the main file:
python main.py
Wait a moment and the results will be displayed in the console.