Code (pytorch) for 'Source Free Unsupervised Domain Adaptation through Semantic Consistency of Hyper-nearest Neighborhood' on Office-31, Office-Home, VisDA-C. This paper has been submitted to IJCAI2021.
You need to download the Office-31, Office-Home, VisDA-C dataset, modify the path of images in each '.txt' under the folder './data/'.
The experiments are conducted on one GPU (NVIDIA RTX TITAN).
- python == 3.7.3
- pytorch ==1.6.0
- torchvision == 0.7.0
- First training model on the source data, Office-31 dataset is shown here.
~/anaconda3/bin/python schn_source.py --trte val --output ckpsglx2020r0/source/ --da uda --gpu_id 0 --dset VISDA-C --net resnet101 --lr 1e-3 --max_epoch 10 --s 0 --seed 2020
- Then adapting source model to target domain, with only the unlabeled target data.
~/anaconda3/bin/python schn_target.py --cls_par 0.2 --da uda --dset VISDA-C --gpu_id 0 --s 0 --output_src ckpsglx2020r0/source/ --output ckpsglx2020r0/target_glx/ --net resnet101 --lr 1e-3 --seed 2020
The results of SCHN is display under the folder './result/'.
The codes are based on SHOT (ICML 2020, also source-free).