[Challenge] [Leaderboard] [Paper]
Team: CASIA-TIM (Members: Lijun Sheng, Zhengbo Wang, Jian Liang)
|–– readme.md
|–– data_list/
| |–– UNIDA/
| | |–– usa_train.txt
| | |–– asia_train.txt
| | |–– asia_test.txt
| | |–– test.txt
| |–– OBJ/
| |–– PLACE/
|
|–– main_unida.py
|–– main_places.py
|–– main_imnet.py
|–– data_list.py
|–– network.py
- python == 3.10.6
- torch ==1.12.0
- torchvision == 0.13.0
- numpy, scipy, sklearn, PIL, argparse
We use the dataset provided by the challenge to generate txt files and place them in the data_list folder according to the names of each dataset (i.e., UNIDA, OBJ, PLACE). If you want to run the code, please modify the absolute paths in all files under data_list folder.
We integrate the source model training, model adaptation, and test file generation in single python code. The test file of the source model is saved as source_test.txt, and the test file based on the adaptive model is saved as target_test.txt.
python main_unida.py --dset UNIDA --gpu_id 0 python main_imnet.py --dset OBJ --gpu_id 1 python main_place.py --dset PLACE --gpu_id 2 If you find this code useful for your research, please cite our paper
@misc{sheng2023self,
title={Self-training solutions for the ICCV 2023 GeoNet Challenge},
author={Sheng, Lijun and Wang, Zhengbo and Liang, Jian},
year={2023}
}