Paper - SALAD: Source-free Active Label Agnostic Domain Adaptation (WACV 2023)

Please cite our paper if you find it useful.

@inproceedings{kothandaraman2023salad,
  title={SALAD: Source-free Active Label-Agnostic Domain Adaptation for Classification, Segmentation and Detection},
  author={Kothandaraman, Divya and Shekhar, Sumit and Sancheti, Abhilasha and Ghuhan, Manoj and Shukla, Tripti and Manocha, Dinesh},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={382--391},
  year={2023}
}

The code structure is as follows:

Active Learning heuristics:
AL/badge_full.py - BADGE active learning sampler
AL/entropy.py - Entropy based active learning sampler
AL/gradient.py - Gradient based active learning sampler

Model folder:
model/attention.py - contains code for guided spatial attention and guided channel attention modules
model/deeplab_multi.py - contains code for the entire SALAD model for semantic segmentation using DeepLab backbone architecture
model/sfda_net.py - contains code for the modulation network

train.py - training script for our model for the task of semantic segmentation
eval_segmentation.py - evaluation script for our model for thetask of semantic segmentation
al_sampler.py - script for choosing samples using our active learning algorithm