AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

CNS-Bench: Benchmarking Image Classifier Robustness Under Continuous Nuisance Shifts

Olaf Dünkel, Artur Jesslen, Jiahao Xie, Christian Theobalt, Christian Rupprecht, Adam Kortylewski

ICCV 2025 · Label-efficient and robust learning › OOD and anomaly detection · approach: Uncertainty and misclassification detection · 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
use textual-inversion IN* distribution instead of plain Stable Diffusion prompts (ablation)image generation · generated images vs ImageNetFID to ImageNet: 33.8 -> 27.1Improved
use IN* instead of plain SD (ablation)image generation · generated images vs ImageNetResNet-50 accuracy: 0.68 -> 0.74Improved
use proposed OOC filter instead of CLIP-based filter (ablation)OOC filtering · manually labeled setfilter accuracy: 0.65 -> 0.88Improved
use proposed OOC filter instead of CLIP-based filter (ablation)OOC filtering · manually labeled setFPR: 0.36 -> 0.12Improved
use proposed OOC filter instead of CLIP-based filter (ablation)OOC filtering · manually labeled setTPR: 0.90 -> 0.88No clear effect
switch from supervised ViT to VMambaclassifier architecture · CNS-BenchrCE: 0.926 -> 0.574Improved
use DeiT3 modern recipe instead of ViT-Bclassifier · CNS-BenchrCE: 0.926 -> 0.610Improved
use DINOv1 self-supervised pre-training instead of supervised IN1kViT-B · CNS-BenchrCE: 0.926 -> 0.636Improved
pre-train on IN21k instead of IN1k (supervised)ViT-B · CNS-BenchrCE: 0.926 -> 0.722Improved
increase DeiT3 size from small to largeDeiT3 · CNS-BenchrCE: 0.747 -> 0.574Improved
increase DeiT3 size from large to hugeDeiT3 · CNS-BenchrCE: 0.574 -> 0.582No clear effect
replace discriminative classifier with diffusion classifierDiT classifier · snow and cartoon shiftsaverage accuracy drop: 0.05 -> 0.106Hurt

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