search

Zhongying-Deng / DIDA

Pytorch implementation for "Dynamic Instance Domain Adaptation" (DIDA-Net, accepted to IEEE T-IP).
10 stars 1 forks source link

Detailed operations for dataset of officeHome #2

Open 23425Ning opened 2 months ago

23425Ning commented 2 months ago

Hello! Thanks for ur excellent work! But i did not see any introductions experiments on dataset of officeHome on this github project which u did in ur paper. So I'd ask could u offer some more detailed operations on how to run ur code on this dataset? Thanks a lot

Zhongying-Deng commented 2 months ago

Many thanks for your interest in our work. The config files for OfficeHome have been updated, please find them via configs/datasets/da and configs/trainers/da/didanet. The backbone used for OfficeHome can be found under dassl/modeling/backbone (please see the latest updated files for the backbone). For training, please follow the following bash script

CUDA_VISIBLE_DEVICES=0 python tools/train.py --root ${YOUR_DATA_PATH} --trainer FixMatch \
    --source-domains art --target-domains clipart \
    --dataset-config-file configs/datasets/da/office_home.yaml --config-file configs/trainers/da/didanet/office_home_dida_single_src.yaml \
    --output-dir output/didanet_office_home/art2clipart \
    MODEL.BACKBONE.NAME resnet50_dida_mixstyle TRAIN.CHECKPOINT_FREQ 10

I am sorry that I have not debugged these new files, so please feel free to let me know if there are any issues with the experiments on OfficeHome.