AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

D-Attn: Decomposed Attention for Large Vision-and-Language Model

Chia-Wen Kuo, Sijie Zhu, Fan Chen, Xiaohui Shen, Longyin Wen

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add V2V diagonalization (ablation)D-Attn (Mistral 0.3 7B) · visual tokens processed in trainingmax |V|: 9k -> 74kImproved
add V2V diagonalization (ablation)D-Attn (Mistral 0.3 7B) · training speedsec/it: 11.25 -> 2.24Improved
add V2V diagonalization alone (ablation)D-Attn (Mistral 0.3 7B) · GQAaccuracy: 61.8 -> 63.4Improved
add V2V diagonalization alone (ablation)D-Attn (Mistral 0.3 7B) · MMStaraccuracy: 36.8 -> 32.6Hurt
add V2V diagonalization alone (ablation)D-Attn (Mistral 0.3 7B) · MMEperception score: 1533.1 -> 1507.6Mixed
add debiased positional encodings on top of diagonalization (ablation)D-Attn (Mistral 0.3 7B) · MMEperception score: 1507.6 -> 1598.6Improved
add debiased positional encodings on top of diagonalization (ablation)D-Attn (Mistral 0.3 7B) · MMStaraccuracy: 32.6 -> 38.3Improved
add debiased positional encodings on top of diagonalization (ablation)D-Attn (Mistral 0.3 7B) · VQA-Taccuracy: 63.4 -> 68.3Improved
use alpha-weighting instead of Tanh gate merging (ablation)D-Attn (Mistral 0.3 7B) · MMEperception score: 1337.6 -> 1598.6Improved
use alpha-weighting instead of Sigmoid gate merging (ablation)D-Attn (Mistral 0.3 7B) · MMStaraccuracy: 35.8 -> 38.3Improved
use alpha-weighting instead of Cascade merging (ablation)D-Attn (Mistral 0.3 7B) · MMEperception score: 1586.1 -> 1598.6Improved
replace self-attention with D-AttnGemma 2 9B LVLM · MMVPaccuracy: 29.3 -> 45.3Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy