Modeling sparse visual state updates instead of full images cuts generated visual tokens ~55.6% and improves interleaved multimodal reasoning over full-image ULMMs.
Title resolution pending
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
-
DeltaV: Thinking with Visual State Updates in Unified Large Multimodal Models
Modeling sparse visual state updates instead of full images cuts generated visual tokens ~55.6% and improves interleaved multimodal reasoning over full-image ULMMs.