AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Two Birds, One Projection: Harmonizing Safety and Utility in LVLMs via Inference-time Feature Projection

Yewon Han, Yumin Seol, Minsoo Jo, EunGyung Kong, Taesup Kim

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 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
apply TBOP null-space projectionLLaVA-1.6-7B · MM-SafetyBench SD-TYPOASR (%): 38.86 -> 5.09Improved
apply TBOP null-space projectionLLaVA-1.6-7B · MMEscore: 1715.13 -> 1870.17Improved
apply TBOP null-space projectionLLaVA-1.6-7B · ScienceQAaccuracy: 57.20 -> 63.10Improved
apply TBOP null-space projectionLLaVA-1.6-13B · MM-Vetscore: 45.63 -> 49.98Improved
apply TBOP null-space projectionLLaVA-1.6-7B · HADESASR (%): 39.06 -> 5.48Improved
apply detect-then-detoxify defenses such as Immune or Protector instead of TBOPLLaVA-1.6-7B · MMEscore: 1715.13 -> 725.39Hurt
add TBOP on top of ECSOECSO · FigStep with SI-Attack (7B)ASR (%): 16.8 -> 5.6Improved
add TBOP on top of ETA (13B)ETA · FigStep with SI-Attack (13B)ASR (%): 0.8 -> 0.6Improved
use MM-SafetyBench-only anchor instead of mixed anchor (ablation)TBOP · MM-SafetyBench SD-TYPOASR (%): 5.09 -> 7.10Hurt
use MME-only anchor instead of mixed anchor (ablation)TBOP · MM-SafetyBench SD-TYPOASR (%): 5.09 -> 36.54Hurt
use a blank dummy image instead of Gaussian noise (ablation)TBOP · MM-SafetyBench SD-TYPOASR (%): 10.12 -> 5.09Improved
use a blank dummy image instead of uniform noise (ablation)TBOP · MM-SafetyBench SD-TYPOASR (%): 10.26 -> 5.09Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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