AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Language Models Can Explain Visual Features via Steering

Javier Ferrando, Enrique Lopez-Cuena, Pablo Agustin Martin-Torres, Daniel Hinjos, Anna Arias-Duart, Dario Garcia-Gasulla

CVPR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded visual concept learning · 2 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
condition Top-k explainer on SAE-steering intervention (Steering-informed Top-k)Top-k explanations, Gemma 3 27B · Middle-layer SAE, ImageNet test (heatmaps)Synthetic activation score: 0.364 -> 0.505Improved
condition Top-k explainer on SAE-steering interventionTop-k explanations, Gemma 3 27B · Middle-layer SAE, ImageNet test (heatmaps)AUROC: 0.791 -> 0.838Improved
condition Top-k explainer on SAE-steering interventionTop-k explanations, InternVL3-14B · Middle-layer SAE, ImageNet test (masks)Synthetic activation score: 0.187 -> 0.254Improved
condition Top-k explainer on SAE-steering interventionTop-k explanations, Gemma 3 27B · Middle-layer SAE, ImageNet test (masks)CLIP score: 0.190 -> 0.193No clear effect
steering-only explanation instead of Top-kGemma 3 27B · Middle-layer SAE, ImageNet testAUROC: 0.723 -> 0.675 (masks)Hurt
steering-only explanation instead of Top-kInternVL3-14B · Middle-layer SAE, ImageNet testIoU score: 0.224 -> 0.220 (masks)No clear effect
scale explainer LM from 4B to 12B to 27B (ablation)Steering, Gemma 3 · Gemma 3 vision encoder SAEMean synthetic activation score and AUROCImproved
filter Top-k explanations by similarity to Steering explanations (ablation)Top-k, Gemma 3 and InternVL3 · Middle-layer SAENormalized synthetic activation score and AUROCImproved
filter Top-k by Steering similarity (ablation)Top-k, Gemma 3 · Middle-layer SAECLIP scoreNo clear effect
use steering instead of top-activating imagesExplanation of background features, Gemma 3 · 300 features with largest IoU gap (heatmaps)% explanations categorized animal: 47.7% -> 0.0%Improved
add steering to Top-kTop-k, Gemma 3 · 300 features with largest IoU gap (heatmaps)% animal explanations (contextual bias): 47.7% -> 38.6%Improved
increase evaluation set size for Top-k (ablation)Top-k explanations · ImageNet test, 1k to 50k imagesSynthetic scores, IoU, CLIPMixed

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