AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

LeGrad: An Explainability Method for Vision Transformers via Feature Formation Sensitivity

Walid Bousselham, Angie Boggust, Sofian Chaybouti, Hendrik Strobelt, Hilde Kuehne

ICCV 2025 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 43 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 ReLU on gradients (ablation)LeGrad (L/14, last layer) · ImageNet-val perturbation negativeAUC: 47.81 -> 49.32Improved
add ReLU on gradients (ablation)LeGrad (L/14, all layers) · ImageNet-val perturbation negativeAUC: 52.01 -> 54.48Improved
add ReLU on gradients (ablation)LeGrad (L/14, all layers) · ImageNet-val perturbation positiveAUC (lower better): 16.80 -> 15.23Improved
use intermediate tokens from all layers (ablation)LeGrad (L/14, with ReLU) · ImageNet-val perturbation negativeAUC: 49.32 -> 54.48Improved
use intermediate tokens from all layers (ablation)LeGrad (L/14, with ReLU) · ImageNet-val perturbation positiveAUC (lower better): 19.95 -> 15.23Improved
use intermediate tokens from all layers (ablation)LeGrad (H/14, with ReLU) · ImageNet-val perturbation positiveAUC (lower better): 19.50 -> 18.26Improved
include [CLS] token in gradient computation (ablation)GradCAM (ViT-B/16) · OpenImagesV7mIoU: 8.18 -> 8.72Improved
add LeGrad-style multi-layer aggregation (ablation)GradCAM (ViT-B/16) · OpenImagesV7mIoU: 8.72 -> 9.51Improved
add LeGrad-style multi-layer aggregation (ablation)GradCAM (ViT-B/16) · ImageNet-val perturbation negativeAUC: 45.26 -> 45.31No clear effect
use LeGrad instead of TextSpanOpenCLIP ViT-B/16 · ImageNet-val perturbation negative targetAUC: 52.81 -> 52.27Hurt
apply LeGrad to ImageBindImageBind · ADE20K Sound PromptedmAP: 18.3 -> 48.0Improved
apply LeGrad to ImageBindImageBind · ADE20K Speech PromptedmAP: 20.2 -> 23.3Improved

Later papers that compared with this method

1 later accepted paper reports LeGrad as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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