SSeg: Active Sparse Point-Label Augmentation for Semantic Segmentation

View PDF HTML (experimental)

Abstract:Semantic segmentation is essential for automating remote sensing analysis in fields like ecology. However, fine-grained analysis of complex aerial or underwater imagery remains an open challenge, even for state-of-the-art models. Progress is frequently hindered by the high cost of obtaining the dense, expert-annotated labels required for model supervision. While sparse point-labels are easier to obtain, they introduce challenges regarding which points to annotate and how to propagate the sparse information. We present SSeg, a novel framework that addresses both issues. SSeg first employs an active sampling strategy to guide annotators, maximizing the value of their point labels. Then, it propagates these sparse labels with a hybrid approach leveraging both the best of SAM2 and superpixel-based methods. Experiments on two diverse monitoring datasets demonstrate SSeg's benefits over state-of-the-art approaches. Our main contribution is a simple but effective interactive annotation tool integrating our algorithms. It enables ecology researchers to leverage foundation models and computer vision to efficiently generate high-quality segmentation masks to process their data.

Subjects:

Computer Vision and Pattern Recognition (cs.CV)

Cite as: arXiv:2510.10163 [cs.CV]

(or arXiv:2510.10163v2 [cs.CV] for this version)

https://doi.org/10.48550/arXiv.2510.10163

arXiv-issued DOI via DataCite

Submission history

From: César Borja [view email] [v1] Sat, 11 Oct 2025 10:56:48 UTC (21,998 KB) [v2] Fri, 27 Mar 2026 08:52:45 UTC (20,605 KB)