Active Sparse Labeling
Study on annotation cost reduction for video understanding tasks
We focus on reducing the annotation cost for video action detection which requires costly frame-wise dense annotations. We study novel active learning (AL) strategies for efficient labeling in videos in our works. First we propose Active Sparse Labeling (ASL) AL strategy to select few high utility frames for all training videos. In our second work, we propose a hybrid Clustering-Aware Uncertainty Scoring (CLAUS) AL strategy to select frames from only portion of training videos with high usefulness. We also provide methods to train video action detection models successfully with limited sparse frame annotations. We provide in-depth details for each of our work here.
Aayush J Rana, Yogesh S Rawat
NeurIPS 2022
Aayush J Rana, Yogesh S Rawat
CVPR 2023
Bibtex
@inproceedings{rana2022are,
title={Are all Frames Equal? Active Sparse Labeling for Video Action Detection},
author={Rana, Aayush J and Rawat, Yogesh S},
booktitle={Advances in Neural Information Processing Systems},
year={2022}
}
@inproceedings{rana2023hybrid,
title={Hybrid Active Learning via Deep Clustering for Video Action Detection},
author={Rana, Aayush J and Rawat, Yogesh S},
booktitle={IEEE/CVF Computer Vision and Pattern Recognition (CVPR) Conference},
year={2023}
}
Team
Aayush J Rana
CRCV, UCF
Yogesh S Rawat
CRCV, UCF