AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

ScAle: Attention Head Scaling as a Minimal Adapter for Spatial Reasoning in Vision–Language Models

Rahul Chowdhury, Timothy Rupprecht, Xuan Shen, Pu Zhao, Yanzhi Wang

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 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
learn bounded per-head attention scalars on a frozen backboneLLaVA-7B · SpatialEval Maze-Navaccuracy: 31.7 -> 68.9Improved
scale attention heads instead of only MLP outputs (ablation)ScAle LLaVA-7B · SpatialEval Maze-Navaccuracy: 44.0 -> 68.9Improved
add MLP scaling on top of attention scaling (ablation)ScAle (attn) LLaVA-7B · SpatialEvalaccuracy: Maze-Nav 68.9 -> 69.7; Spatial-Map 45.0 -> 45.3No clear effect
increase training data from 5% to 20% (ablation)ScAle (attn) LLaVA-7B · SpatialEval Spatial-Gridaccuracy: 51.7 -> 68.6Improved
combine LoRA on QKV with ScAle scalingLoRA · SpatialEval Maze-Navaccuracy: 71.6 -> 78.9 (vs LoRA-all)Improved
apply LoRA only to last layer instead of ScAle attention scalingLLaVA-7B · SpatialEval Spatial-Gridaccuracy: 68.6 -> 29.2Hurt
train ScAle on Spatial-Grid and evaluate on real-world VGQAQwen2.5-VL-3B · VGQAaccuracy: 84.5 -> 92.2Improved
train ScAle on spatial task, evaluate on other spatial taskQwen2.5-VL-3B · Maze-Navaccuracy: 26.0 -> 27.3 (Spatial-Grid adapter)No clear effect
train ScAle on spatial tasks, evaluate on GLUE/SST2Qwen2.5-VL-3B · GLUE/SST2accuracy: 65.8 -> 58.8 (Maze-Nav), 88.3 (Spatial-Map)Mixed
apply ScAle to hallucination benchmarkLLaVA-7B · POPEaccuracy: 82.1 -> 88.2Improved

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