A training-free two-stage token pruning method for video-language models, using eigenvalue decay of token correlations to set a content-adaptive retention ratio.
Divprune: Diversity-based visual token pruning for large multimodal models
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
-
Adaptive Two-Stage Visual Token Pruning for Efficient Inference in Video-Language Models
A training-free two-stage token pruning method for video-language models, using eigenvalue decay of token correlations to set a content-adaptive retention ratio.