VASparse combines visual-aware token pruning, embedding-based visual contrastive decoding, and an attention-sink penalty to reduce visual hallucinations in LVLMs without extra training.
Steven Adler
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
VASparse: Towards Efficient Visual Hallucination Mitigation via Visual-Aware Token Sparsification
VASparse combines visual-aware token pruning, embedding-based visual contrastive decoding, and an attention-sink penalty to reduce visual hallucinations in LVLMs without extra training.