- pytorch = 1.0.1
- torchvision = 0.2.1
- numpy = 1.17.2
- pillow = 6.2.0
- python3.6
- cuda10
The following are the commands for each task. Here, wt represents the parameter for weight of the transfer loss.
Office-31
python train.py --gpu_id 0 --dset office --s_dset_path data/office/amazon_list.txt --t_dset_path data/office/dslr_list.txt --output_dir chatty/adn --wt 0.001 --domains A_to_D
Office-Home
python train.py --gpu_id 0 --dset office-home --s_dset_path data/office-home/Art.txt --t_dset_path data/office-home/Clipart.txt --output_dir chatty/ArCl --wt 0.0001 --domains Ar_to_Cl
Fhist
python train.py --gpu_id 0 --dset fhist --s_dset_path data/fhist/labeled_source.txt --t_dset_path data/fhist/unlabeled_target.txt --output_dir chatty/CrcNct --wt 0.0001 --domains CRC_to_NCT
The codes are heavily borrowed from GVB