AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

VISOR++ : VISUAL INPUT BASED STEERING FOR LARGE VISION LANGUAGE MODELS

Ravikumar Balakrishnan, Mansi Phute

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 0 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
use ensemble VISOR++ image instead of system promptLLaVA-1.5-7B · Refusal (negative steering)BAS: 0.698 -> 0.353Improved
use ensemble VISOR++ image instead of system promptIDEFICS2-8B · Sycophancy (negative steering)BAS: 0.759 -> 0.581Improved
optimize per-model VISOR++ image instead of steering vectorIDEFICS2-8B · Refusal (positive steering)BAS: 0.817 -> 0.94Improved
jointly optimize one ensemble image instead of per-model imagesLLaVA-1.5-7B · Sycophancy (negative steering)BAS: 0.393 -> 0.623Hurt
jointly optimize one ensemble image instead of per-model imagesIDEFICS2-8B · Refusal (negative steering)BAS: 0.231 -> 0.29No clear effect
insert ensemble VISOR++ image (trained on LLaVA and IDEFICS2) on unseen modelGPT-4-Turbo · Survival instinct (negative steering)behavior fraction: 0.388 -> 0.312Improved
insert ensemble VISOR++ image on unseen modelGPT-4-Turbo · Sycophancy (negative steering)behavior fraction: 0.460 -> 0.390Improved
insert ensemble VISOR++ image on unseen modelClaude Sonnet 3.5 · Refusal (negative steering)behavior fraction: 0.609 -> 0.609No clear effect
insert ensemble VISOR++ image on unseen modelQwen2-vl-7b · Refusal (negative steering)BAS: 0.866 -> 0.859No clear effect
insert ensemble VISOR++ image on unseen modelLlama-3.2-11B · Sycophancy (negative steering)BAS: 0.496 -> 0.518Hurt
insert ensemble VISOR++ image on unseen modelLLaVA-NeXT · Refusal (negative steering)BAS: 0.879 -> 0.852Improved
add ensemble VISOR++ image versus random imageLLaVA-1.5-7B · MMLU (14,000 samples)mean score: 0.491 -> 0.492No clear effect

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