Contamination Map Methodology

Vision / Model

MuSc

Organisation
Huazhong University of Science and Technology et al.
Family
MuSc
Released
2024-01-30 (arXiv v1 submission date (https://arxiv.org/abs/2401.16753))
Weights
Unknown · MIT
Links
PaperCode

Training data disclosure: Full list

Level determined for 0% of benchmark cells.

“We leverage such a discriminative characteristic to design a novel zero-shot AC/AS method by Mutual Scoring (MuSc) of the unlabeled images, which does not need any training or prompts.”
raw.githubusercontent.com · readme · Introduction · retrieved 2026-10-15
“Our MuSc is a pure zero-shot method that does not require any labeled images.”
arxiv.org · paper · Section 4 · retrieved 2026-10-15

Built on

Everything these base models may have seen is inherited by MuSc.

Benchmark cells

Industrial anomaly detection

Unknown MVTec AD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown MVTec AD 2 · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown MVTec LOCO AD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown VisA · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown Real-IAD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown MPDD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown BTAD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Unknown KolektorSDD2 · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail

Notes

Training-free and transductive: at inference, the unlabelled TEST images of the target dataset score each other (Mutual Scoring Mechanism), and RsCIN re-scores image-level results using the image-level neighbourhood over the test set. No parameters are fitted, so no training entry; but results depend on the whole unlabelled test set being available ('we propose the first method that only uses the unlabeled test images'). Code also supports OpenCLIP LAION-400M, DINO and DINOv2 backbones (paper appendix/README ablations); the reported default is OpenAI CLIP ViT-L/14-336.

Sources