tntek / gkd

Code (pytorch) for 'Model Adaptation through Hypothesis Transfer with Gradual Knowledge Distillation' that has been accepted by IROS2021.

GKD

Code (pytorch) for 'Model Adaptation through Hypothesis Transfer with Gradual Knowledge Distillation' on Office-31, Office-Home, VisDA-C. This paper has been accepted by IROS2021.

Preliminary

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

Training and evaluation

1. First training model on the source data, Office-Home dataset is shown here.

~/anaconda3/bin/python gkd_source.py --trte val --output ckpsmix2020r0/source/ --da uda --gpu_id 0 --dset office-home --max_epoch 50 --s 0 --seed 2020

2. Then adapting source model to target domain, with only the unlabeled target data.

~/anaconda3/bin/python gkd_target.py --cls_par 0.05 --da uda --dset office-home --gpu_id 0 --s 0 --t 1 --output_src ckpsmix2020r0/source/ --output ckpsmix2020r0/target_mix/ --seed 2020

Results

The results of GKD is display under the folder './result/'.

Acknowledgement

The codes are based on SHOT (ICML 2020, also source-free).

Contact

• [email protected]