pip install -r requirements.txt
Because the datasets are quite large, we have only uploaded Indian Pines to the dataset folder, the others can be found via the citations in our paper or by request to zhoufangqin123@gmail.com.
To run the search scripts, specify a search proxy in the search_xx.sh, then run with bash command, for example,
$bash search_indian_sf_final.sh
To run the retrain scripts, specify a dataset and path in the train_xx.sh, then run with the bash command,
$bash train_searched_sf.sh
Individual retrain and search results are saved under outputs/
Merged results of each dataset are saved under results/
For visualizing results, see analysis.ipynb
If you have any questions, please contact zhoufangqin123@gmail.com
The implementation is built upon TF_TAS
Thanks for their great work!