CSteer steers general LMMs for multi-region visual referring via pre-computed contextual vectors and inference-time representation editing, outperforming specialized models on benchmarks.
# Output Format: IMPORTANT: Output ONLY the rewritten response text
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
-
Referring Multiple Regions with Large Multimodal Models via Contextual Latent Steering
CSteer steers general LMMs for multi-region visual referring via pre-computed contextual vectors and inference-time representation editing, outperforming specialized models on benchmarks.