RoRA allocates a fixed visual-token budget into semantic core, context, and detail roles guided by attention-anchored regions, improving pruned MLLM accuracy and speed without training.
Proceedings of the AAAI Conference on Artificial Intelligence , volume=
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
-
RoRA: Role-Oriented Regional Allocation for Visual Token Pruning in MLLMs
RoRA allocates a fixed visual-token budget into semantic core, context, and detail roles guided by attention-anchored regions, improving pruned MLLM accuracy and speed without training.