Guided SAM: Label-Efficient Part Segmentation
S.B. van Rooij, G.J. Burghouts · 2025
Read Guided SAM: Label-Efficient Part Segmentation by S.B. van Rooij, G.J. Burghouts (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-07434.txt.
Description
Van Rooij and Burghouts cut the labeling cost of teaching software to outline object parts, in a 2025 paper. Finding parts, not just whole objects, matters for recognition and for robots that must grip things.
The base tool they use, Segment Anything, needs a person to point at each part, and it tends to spill beyond part borders. The authors train simple patch classifiers from coarse, cheap labels. Patches that match a part are merged into regions, which then prompt the base tool automatically. On a car parts data set, the overlap score rose from 0.37 to 0.49. The labels needed were about five times cheaper to make than full outlines. The method suits fields where expert labeling is the bottleneck.
Catalog record
| CREATOR | S.B. van Rooij, G.J. Burghouts |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-07434