Contamination MapVision / Model
InCTRL, trained on MVTec AD
- Organisation
- Singapore Management University (Jiawen Zhu, Guansong Pang)
- Family
- InCTRL
- Released
- 2024-03-11 (arXiv v1 submission date (https://arxiv.org/abs/2403.06495))
- Checkpoints
- Google Drive 'Pre-train Models' folder (file names not verified)
- Weights
- Open · Apache-2.0
- Links
- PaperCode
Training data disclosure: Full list
Level determined for 62% of benchmark cells.
- TestTest data seen1
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen0
- NoNot seen4
- ?Unknown3
“the full dataset of MVTec AD, including both training set and test set, is used as the auxiliary training data, on which AD models are trained, and they are subsequently evaluated on the test set of the other eight datasets without any further training.”
Built on
Everything these base models may have seen is inherited by InCTRL, trained on MVTec AD.
-
OpenCLIP (LAION-400M checkpoints)
· disclosure: Full list
Evidence
“the same CLIP implementation, OpenCLIP [24], and its public pre-trained backbone ViT-B/16+ in our experiments.”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 “_C.TRAIN.CHECKPOINT_FILE_PATH = "vit_b_16_plus_240-laion400m_e32-699c4b84.pt"”
raw.githubusercontent.com · code · open_clip/config/defaults.py · retrieved 2026-10-15
Other declared training data
-
MVTec AD · split: full · fine-tuning
Evidence
“To assess the GAD performance, MVTec AD, the combination of its training and test sets, is used as the auxiliary training data”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 “When training GAD models with MVTec AD or VisA datasets, we utilize their complete datasets, including both training and test data.”
arxiv.org · paper · Table 4 caption · retrieved 2026-10-15
Benchmark cells
Industrial anomaly detection
Test data seen MVTec AD · in fine-tuning data: the whole dataset was used
Evidence (2)
“To assess the GAD performance, MVTec AD, the combination of its training and test sets, is used as the auxiliary training data”
“When training GAD models with MVTec AD or VisA datasets, we utilize their complete datasets, including both training and test data.”
Not seen MVTec AD 2 · absent from the complete list of training data and from its base models' data
Evidence (6)
“the full dataset of MVTec AD, including both training set and test set, is used as the auxiliary training data, on which AD models are trained, and they are subsequently evaluated on the test set of the other eight datasets without any further training.”
“the same CLIP implementation, OpenCLIP [24], and its public pre-trained backbone ViT-B/16+ in our experiments.”
“_C.TRAIN.CHECKPOINT_FILE_PATH = "vit_b_16_plus_240-laion400m_e32-699c4b84.pt"”
“The image-text-pairs have been extracted from the Common Crawl web data dump and are from random web pages crawled between 2014 and 2021.”
“arXiv:2503.21622v1 [cs.CV] 27 Mar 2025”
“All image data is available at https://www.mvtec.com/company/research/datasets/mvtec-ad-2.”
Not seen MVTec LOCO AD · absent from the complete list of training data and from its base models' data
Evidence (5)
“the full dataset of MVTec AD, including both training set and test set, is used as the auxiliary training data, on which AD models are trained, and they are subsequently evaluated on the test set of the other eight datasets without any further training.”
“the same CLIP implementation, OpenCLIP [24], and its public pre-trained backbone ViT-B/16+ in our experiments.”
“_C.TRAIN.CHECKPOINT_FILE_PATH = "vit_b_16_plus_240-laion400m_e32-699c4b84.pt"”
“The image-text-pairs have been extracted from the Common Crawl web data dump and are from random web pages crawled between 2014 and 2021.”
“Published: 22 February 2022”
Not seen VisA · absent from the complete list of training data and from its base models' data
Evidence (6)
“the full dataset of MVTec AD, including both training set and test set, is used as the auxiliary training data, on which AD models are trained, and they are subsequently evaluated on the test set of the other eight datasets without any further training.”
“the same CLIP implementation, OpenCLIP [24], and its public pre-trained backbone ViT-B/16+ in our experiments.”
“_C.TRAIN.CHECKPOINT_FILE_PATH = "vit_b_16_plus_240-laion400m_e32-699c4b84.pt"”
“The image-text-pairs have been extracted from the Common Crawl web data dump and are from random web pages crawled between 2014 and 2021.”
“arXiv:2207.14315v1 [cs.CV] 28 Jul 2022”
“We release the Visual Anomaly (VisA) Dataset consisting of 10,821 high-resolution color images (9,621 normal and 1,200 anomalous samples) covering 12 objects in 3 domains”
Not seen Real-IAD · absent from the complete list of training data and from its base models' data
Evidence (5)
“the full dataset of MVTec AD, including both training set and test set, is used as the auxiliary training data, on which AD models are trained, and they are subsequently evaluated on the test set of the other eight datasets without any further training.”
“the same CLIP implementation, OpenCLIP [24], and its public pre-trained backbone ViT-B/16+ in our experiments.”
“_C.TRAIN.CHECKPOINT_FILE_PATH = "vit_b_16_plus_240-laion400m_e32-699c4b84.pt"”
“The image-text-pairs have been extracted from the Common Crawl web data dump and are from random web pages crawled between 2014 and 2021.”
“arXiv:2403.12580v1 [cs.CV] 19 Mar 2024”
Unknown MPDD · unknown for its base model OpenCLIP (LAION-400M checkpoints): training data not disclosed in enough detail
Unknown BTAD · unknown for its base model OpenCLIP (LAION-400M checkpoints): training data not disclosed in enough detail
Unknown KolektorSDD2 · unknown for its base model OpenCLIP (LAION-400M checkpoints): training data not disclosed in enough detail
Notes
Frozen OpenCLIP ViT-B/16+ (LAION-400M) with a trained adapter and residual head. Trained on the whole MVTec AD dataset (train + test) and evaluated on the other eight datasets (VisA, ELPV, SDD, AITEX, BrainMRI, HeadCT, MNIST, CIFAR-10). Few-shot normal prompts at inference are sampled from the target dataset's training set. A third-party paper (AdaptCLIP) notes InCTRL provides separate official models for different shot numbers (2/4/8).
Sources
- Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Prompts (CVPR 2024) · paper · retrieved 2026-10-15
- InCTRL README · readme · retrieved 2026-10-15
- open_clip/config/defaults.py · code · retrieved 2026-10-15