AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Lightweight Structure-Aware Attention for Visual Understanding

Heeseung Kwon, Francisco M. Castro, Manuel Jesús Marín-Jiménez, Nicolás Guil, Karteek Alahari

IJCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 1 citations (OpenAlex)

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 structural patterns (Eq.5 instead of Eq.4) (ablation)structure-aware attention · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 72.4 -> 74.2Improved
add Hadamard correlation (ablation)structure-aware attention · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 74.2 -> 74.9Improved
approximate global convolutions with FFT (ablation)structure-aware attention · ImageNet-1K (LiSANet-I, 150 epochs)FLOPs (G): 3.92 -> 1.21Improved
use RPE as multiplicative weights (Eq.4) instead of additive RPE (ablation)self-attention · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 72.2 -> 72.4Improved
use independent RPEs per head (ablation)structure-aware attention · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 72.7 -> 73.6Improved
increase structural pattern size D from 1 to 16 (ablation)LiSA · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 71.3 -> 74.9Improved
use global kernel instead of local 7x7 (ablation)LiSA · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 73.8 -> 74.9Improved
use local 7x7 instead of 3x3 kernel (ablation)LiSA · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 71.9 -> 73.8Improved
replace self-attention with LiSAisotropic ViT · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 71.0 -> 74.9Improved
replace GF layer with LiSAisotropic ViT · ImageNet-1K (LiSANet-I, 150 epochs)top-1 (%): 69.5 -> 74.9Improved
ImageNet-21K pretrainingHyLiSANet-B · ImageNet-1K valtop-1 (%): 85.0 -> 86.2Improved
replace CSwin-T with HyLiSANet-S as backboneMask R-CNN · COCObox AP: 46.7 -> 47.7Improved

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