Contamination Map Methodology

Vision / Model

Anomaly-OneVision (Anomaly-OV)

Organisation
Johns Hopkins University / Honda Research Institute USA
Family
Anomaly-OV
Released
2025-02-11 (arXiv v1 submission date (https://arxiv.org/abs/2502.07601))
Checkpoints
Anomaly-OV-0.5B, Anomaly-OV-7B
Weights
Unknown
Links
PaperCode

Training data disclosure: Partial list

Level determined for 25% of benchmark cells.

“For a unified visual inspection dataset, precise instruction data is collected from MVTec AD [2], the training set of BMAD [1], Anomaly-ShapeNet [50], Real3D-AD [60], and MVTec-3D AD [4]”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15
“This stage utilizes all of the data with anomaly labels in Anomaly-Instruct-125k. Similar to previous works [6, 110], when evaluating the model on the datasets contained in the training set, the corresponding datasets are replaced by VisA [115].”
arxiv.org · paper · Section 4.3 / training · retrieved 2026-10-15

Built on

Everything these base models may have seen is inherited by Anomaly-OneVision (Anomaly-OV).

Other declared training data

Benchmark cells

Industrial anomaly detection

Seen, split unclear MVTec AD · in fine-tuning data: the dataset was used; the source does not say which part

Evidence (2)
“For a unified visual inspection dataset, precise instruction data is collected from MVTec AD [2], the training set of BMAD [1], Anomaly-ShapeNet [50], Real3D-AD [60], and MVTec-3D AD [4]”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15
“This stage utilizes all of the data with anomaly labels in Anomaly-Instruct-125k. Similar to previous works [6, 110], when evaluating the model on the datasets contained in the training set, the corresponding datasets are replaced by VisA [115].”
arxiv.org · paper · Section 4.3 / training · retrieved 2026-10-15

Unknown MVTec AD 2 · unknown for its base model LLaVA-OneVision: training data not disclosed in enough detail

Unknown MVTec LOCO AD · unknown for its base model LLaVA-OneVision: unknown for its base model SigLIP: training data not disclosed in enough detail

Seen, split unclear VisA · in fine-tuning data: the dataset was used; the source does not say which part

Evidence (1)
“This stage utilizes all of the data with anomaly labels in Anomaly-Instruct-125k. Similar to previous works [6, 110], when evaluating the model on the datasets contained in the training set, the corresponding datasets are replaced by VisA [115].”
arxiv.org · paper · Section 4.3 / training · retrieved 2026-10-15

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

Unknown MPDD · unknown for its base model LLaVA-OneVision: unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown BTAD · unknown for its base model LLaVA-OneVision: unknown for its base model SigLIP: training data not disclosed in enough detail

Unknown KolektorSDD2 · unknown for its base model LLaVA-OneVision: unknown for its base model SigLIP: training data not disclosed in enough detail

Notes

MLLM specialist (anomaly expert + LLaVA-OneVision). Zero-shot detection evaluated on MVTec AD, VisA, AITEX, ELPV, BTAD, MPDD, BrainMRI, HeadCT, Br35H. README releases code and the instruction dataset/benchmark (via project page) with training scripts; released weights not confirmed. WebAD (scraped web images) is marked open_ended.

Sources