AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Efficient Unsupervised Shortcut Learning Detection and Mitigation in Transformers

Lukas Kuhn, Sari Sadiya, Jörg Schlötterer, Florian Buettner, Christin Seifert, Gemma Roig

ICCV 2025 · Detection and segmentation › Open-set and few-shot detection · approach: Industrial anomaly detection · 8 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 ASM (patch ablation + last-layer retraining) to the fine-tuned ViTViT-B/16 baseline · ISICworst-group accuracy (%): 51.7 -> 74.7Improved
add ASM to the fine-tuned ViTViT-B/16 baseline · Knee Radiographsworst-group accuracy (%): 37.3 -> 81.7Improved
add ASM to the fine-tuned ViTViT-B/16 baseline · ImageNet-Wworst-group accuracy (%): 69.0 -> 87.0Improved
add last-layer retraining after patch ablation (ablation)ASM · ISICworst-group accuracy (%): 61.0 -> 74.7Improved
add last-layer retraining after patch ablation (ablation)ASM · ImageNet-Wworst-group accuracy (%): 75.0 -> 87.0Improved
add last-layer retraining after patch ablation (ablation)ASM · Knee Radiographsworst-group accuracy (%): 80.0 -> 81.7No clear effect
patch ablation only, no retraining, versus baseline (ablation)ASM w/o retraining · Knee Radiographsworst-group accuracy (%): 37.3 -> 80.0Improved
use unsupervised ASM instead of group-annotated DFRDFR · ISICworst-group accuracy (%): 59.3 -> 74.7Improved
use subsampling groups (SUBG) instead of baselineViT-B/16 baseline · ISICaverage-group accuracy (%): 86.2 -> 74.3Hurt
use 3 clusters instead of 2 (overclustering) (ablation)ASM w/o retraining · ISICworst-group accuracy (%): 61.0 -> 56.7No clear effect
use ASM instead of JTT for runtimeJTT · ISICruntime (s): 2886.3 -> 121.1Improved

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