CG-ICS uses an MLLM to generate candidate textual concepts, scores them with SAM3 in a tree search, and combines the best concept with a visual exemplar to segment the query stably.
Overcoming support dilution for robust few-shot semantic segmentation.arXiv preprint arXiv:2501.13529, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Toward Robust In-Context Segmentation via Concept Guidance
CG-ICS uses an MLLM to generate candidate textual concepts, scores them with SAM3 in a tree search, and combines the best concept with a visual exemplar to segment the query stably.