Contamination Map Methodology

Vision / Model

AdaCLIP, trained on MVTec AD + ColonDB

Organisation
Huazhong University of Science and Technology et al.
Family
AdaCLIP · version pretrained_mvtec_colondb.pth
Released
2024-07-22 (arXiv v1 submission date (https://arxiv.org/abs/2407.15795))
Checkpoints
weights/pretrained_mvtec_colondb.pth
Weights
Open · MIT
Links
PaperCode

Training data disclosure: Full list

Level determined for 12% of benchmark cells.

“The description to the utilized training set in our paper is not accurate. By default, we utilize MVTec AD & ColonDB for training, and VisA & ClinicDB are utilized for evaluations on MVTec AD & ColonDB.”
raw.githubusercontent.com · readme · Corrections · retrieved 2026-10-15
“# pre-trained from MVTec and ColonDB ckt_path="weights/pretrained_mvtec_colondb.pth"”
raw.githubusercontent.com · code · test.sh · retrieved 2026-10-15

Built on

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

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 (2)
“The description to the utilized training set in our paper is not accurate. By default, we utilize MVTec AD & ColonDB for training, and VisA & ClinicDB are utilized for evaluations on MVTec AD & ColonDB.”
raw.githubusercontent.com · readme · Corrections · retrieved 2026-10-15
“Only utilize the test dataset for both training and testing”
raw.githubusercontent.com · code · dataset/base_dataset.py · 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

Hybrid static/dynamic learnable prompts in a frozen CLIP ViT-L/14-336 (OpenAI). Used (test.sh) for all evaluations except MVTec AD and ColonDB: VisA, MPDD, BTAD, SDD (KolektorSDD), DAGM, DTD-Synthetic and medical sets. The paper text says MVTec AD & ClinicDB; the README 'Corrections' section says MVTec AD & ColonDB, which matches the code (test.sh, HF Space app). README training recipe uses the target-side dataset (e.g. VisA) as validation set to pick the best epoch ('choose the best model based on performance on the validation set').

Sources