Region-Based Active Learning for Efficient Labelling in Semantic Segmentation

Abstract

Semantic segmentation requires dense per-pixel annotations, making manual labeling extremely costly. We propose a region-based active learning strategy that selects superpixel regions — rather than individual pixels or entire images — for annotation. By querying the most informative regions across the image pool, our method reduces annotation effort significantly while maintaining segmentation quality, providing a practical path toward efficient labeling for dense prediction tasks.

BibTeX

@inproceedings{kasarla2019region,
  title={Region-Based Active Learning for Efficient Labelling in Semantic Segmentation},
  author={Kasarla, Tejaswi and Nagendar, G and Hegde, Guruprasad and Balasubramanian, Vineeth N and Jawahar, CV},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  year={2019}
}