Our codes are developed based on the codes released for CDAN. (Long M, Cao Z, Wang J, et al. Conditional adversarial domain adaptation[C]//Advances in Neural Information Processing Systems. 2018: 1640-1650.)
Office-31 dataset can be found on https://people.eecs.berkeley.edu/~jhoffman/domainadapt/. After downloading the data set, put the image file under the path "../data/office/domain_adaptation_images/". Data list files named with format "_10_list.txt" should be used as source data. "_11_list.txt" should be used as target data.
Office-Home dataset can be found on http://hemanthdv.org/OfficeHome-Dataset/. After downloading the data set, put the image file under the path "../data/office-home/images/". Data list files named with format "_k.txt" should be used as source data. "_uk.txt" should be used as target data.
Here we give an example of the training command for task D->A:
python train_image.py --gpu_id 0 --net ResNet50 --dset office --test_interval 50 --s_dset_path ../data/office/dslr_10_list.txt --t_dset_path ../data/office/amazon_11_list.txt OSDA
This command can be run directly after decompression.