Contamination Map Methodology

Vision / Model

AnomalyCLIP, trained on MVTec AD

Organisation
Zhejiang University / Singapore Management University et al.
Family
AnomalyCLIP · version checkpoints/9_12_4_multiscale_visa/epoch_15.pth
Released
2023-10-29 (arXiv v1 submission date (https://arxiv.org/abs/2310.18961))
Checkpoints
checkpoints/9_12_4_multiscale_visa/epoch_15.pth
Weights
Open · MIT
Links
PaperCode

Training data disclosure: Full list

Level determined for 12% of benchmark cells.

“Then, we fine-tune AnomalyCLIP using the test data on MVTec AD and evaluate the ZSAD performance on other datasets.”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15
“Main results (We test all datasets by training once on MVTec AD. For MVTec AD, AnomalyCLIP is trained on VisA.)”
raw.githubusercontent.com · readme · Main results · retrieved 2026-10-15

Built on

Everything these base models may have seen is inherited by AnomalyCLIP, trained on MVTec AD.

Other declared training data

Benchmark cells

Industrial anomaly detection

Test data seen MVTec AD · in fine-tuning data: its test split was used for training

Evidence (1)
“We fine-tune AnomalyCLIP on the test data on MVTec AD and test the performance for other datasets.”
arxiv.org · paper · Appendix A.1 · retrieved 2026-10-15

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

Learnable object-agnostic text prompts (CLIP frozen). Trained on the MVTec AD test set (15 epochs) and used for all reported zero-shot results except MVTec AD itself (VisA, MPDD, BTAD, SDD, DAGM, DTD-Synthetic and the medical datasets). In the repo the MVTec-trained checkpoint is the folder named '9_12_4_multiscale_visa' (train.sh: base_dir ..._multiscale_visa with --dataset mvtec; test.sh uses it with --dataset visa); the naming is counter-intuitive. The paper also reports a variant fine-tuned on a medical set built from ColonDB plus the test split of Endo (Table 3); no checkpoint for it is released.

Sources