AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Foveated Reasoning: Stateful, Action-based Visual Focusing for Vision-Language Models

Juhong Min, Lazar Valkov, Vitali Petsiuk, Hossein Souri, Deen Dayal Mohan

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 3 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 coldstart foveation SFT (ablation)Qwen2.5-VL-7B 336^2 · DocVQAAccuracy: 55.1 -> 64.2Improved
add coldstart foveation SFT (ablation)Qwen2.5-VL-7B 336^2 · TextVQAAccuracy: 77.2 -> 82.2Improved
add coldstart foveation SFT (ablation)Qwen2.5-VL-7B 336^2 · Visual7WAccuracy: 46.5 -> 58.3Improved
add RL (lambda_reg 1.0) after coldstart (ablation)FoveateR-7B coldstart · DocVQAAccuracy: 64.2 -> 61.5Hurt
add RL (lambda_reg 1.0) after coldstart (ablation)FoveateR-7B coldstart · TextVQAAccuracy: 82.2 -> 81.2Mixed
add RL (lambda_reg 1.0) after coldstart (ablation)FoveateR-7B coldstart · Visual7WAccuracy: 58.3 -> 57.2Mixed
use action-based single-pass foveation instead of multi-pass VisCoTFoveateR-7B · DocVQA (336^2)Accuracy: 47.6 -> 83.3Improved
use action-based single-pass foveation instead of multi-pass VisCoTFoveateR-7B · Visual CoT (336^2)Avg visual tokens: 1,152 -> 322.5Improved
use FoveateR instead of VisCoTFoveateR-7B · Flickr30k (336^2)Accuracy: 66.8 -> 60.8Hurt
increase low-res input from 224^2 to 336^2FoveateR-7B · DocVQAAccuracy: 73.9 -> 83.3Improved
scale backbone from 3B to 7BFoveateR · DocVQA (336^2)Accuracy: 74.0 -> 83.3Improved
use FoveateR-3B instead of VisCoT-7BFoveateR-3B · V* Bench GPT4V-hardAccuracy: 11.8 -> 76.5Improved

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