The proposed UC-SFDA is programmed in PyTorch 1.9.0 and trained on a server with Intel(R) Core(TM) i9-12900K CPU and NVIDIA GeForce RTX3090 GPU.
The training log file and the model file are available by linking ：https://pan.baidu.com/s/1EhKviVvBHOLKPhysCGYdfg?pwd=SFDA code：SFDA .
Download the pre-trained model from URL https://storage.googleapis.com/vit_models/imagenet21k/R50+ViT-B_16.npz to path “./model/vit_checkpoint/imagenet21k”
- Please manually download the datasets Office, Office-Home, from the official websites, and modify the path of images in each '.txt' under the folder './data/'.
Train model on the source domain(The default task is domain A in OFFICE-31, and the default parameter configuration is the recommended configuration):
python source_pretrain.py
Adaptation to other target domains (The default source domain is A and the target domain is D, and the default parameter configuration is the recommended configuration):
python target_adaptation.py