SAM3 performs poorly with text-only prompts on nuclei, is sensitive to visual prompt quality, gains modestly from few-shot examples but lacks robustness to noise, and shows a large gap to task-specific adapters on NuInsSeg, PanNuke, and GlaS datasets.
Segment anything model (sam) for digital pathology: Assess zero-shot seg- mentation on whole slide imaging,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Is SAM3 ready for pathology segmentation?
SAM3 performs poorly with text-only prompts on nuclei, is sensitive to visual prompt quality, gains modestly from few-shot examples but lacks robustness to noise, and shows a large gap to task-specific adapters on NuInsSeg, PanNuke, and GlaS datasets.