Contamination MapVision / Model
AnomalyGPT, unsupervised on MVTec AD
- Organisation
- Institute of Automation, Chinese Academy of Sciences / Objecteye / Wuhan AI Research
- Family
- AnomalyGPT · version train_mvtec
- Released
- 2023-08-29 (arXiv v1 submission date (https://arxiv.org/abs/2308.15366))
- Checkpoints
- FantasticGNU/AnomalyGPT train_mvtec/pytorch_model.pt
- Weights
- Open
- Links
- PaperCode
Training data disclosure: Full list
Level determined for 12% of benchmark cells.
- TestTest data seen0
- SeenSeen, split unclear0
- TrainTraining data seen1
- Related data seen0
- NoNot seen0
- ?Unknown7
“Consistent with previous IAD methods, we only use the normal data from these datasets for training.”
“Unsupervised on MVTec-AD”
Other declared training data
-
MVTec AD · split: train of industrial-ad · fine-tuning
Code (code/datasets/mvtec.py) keeps only files under train/good.
Evidence
“Consistent with previous IAD methods, we only use the normal data from these datasets for training.”
arxiv.org · paper · Section 4, Datasets · retrieved 2026-10-15 “We primarily adopt the approach proposed by NSA [24] to simulate anomalous data.”
arxiv.org · paper · Section 3.4 · retrieved 2026-10-15 -
pre-training data of PandaGPT (pandagpt4_visual_instruction_data) (not yet matched to a dataset) · split: unknown · fine-tuning
PandaGPT visual instruction data (HF dataset openllmplayground/pandagpt_visual_instruction_dataset); the paper says LLaVA/PandaGPT were pre-trained on 160k images with multi-turn conversations.
Evidence
“We perform alternating training using both the pre-training data of PandaGPT [25] and our anomaly image-text data.”
arxiv.org · paper · Section 4, Implementation Details · retrieved 2026-10-15 -
Pre-training data of base model imagebind-huge (not tracked here) (broad web data: may contain benchmark test sets) · split: unknown
Evidence
“We utilize ImageBind-Huge [8] as the image encoder and Vicuna-7B [3] as the inferential LLM, connected through a linear layer.”
arxiv.org · paper · Section 4, Implementation Details · retrieved 2026-10-15 -
Pre-training data of base model vicuna-7b (not tracked here) (broad web data: may contain benchmark test sets) · split: unknown
Evidence
“We utilize ImageBind-Huge [8] as the image encoder and Vicuna-7B [3] as the inferential LLM, connected through a linear layer.”
arxiv.org · paper · Section 4, Implementation Details · retrieved 2026-10-15 -
Pre-training data of base model pandagpt-7b (not tracked here) (broad web data: may contain benchmark test sets) · split: unknown
Evidence
“We initialize our model using pre-trained parameters from PandaGPT [25].”
arxiv.org · paper · Section 4, Implementation Details · retrieved 2026-10-15 “In our experiments and online demo, we use the Vicuna-7B and `openllmplayground/pandagpt_7b_max_len_1024` due to the limitation of computation resource.”
raw.githubusercontent.com · readme · 2.4 · retrieved 2026-10-15
Benchmark cells
Industrial anomaly detection
Training data seen MVTec AD · in fine-tuning data: only the training split was used, not the benchmark's test data
Evidence (2)
“Consistent with previous IAD methods, we only use the normal data from these datasets for training.”
“We primarily adopt the approach proposed by NSA [24] to simulate anomalous data.”
Unknown MVTec AD 2 · training data not disclosed in enough detail
Unknown MVTec LOCO AD · training data not disclosed in enough detail
Unknown VisA · training data not disclosed in enough detail
Unknown Real-IAD · training data not disclosed in enough detail
Unknown MPDD · training data not disclosed in enough detail
Unknown BTAD · training data not disclosed in enough detail
Unknown KolektorSDD2 · training data not disclosed in enough detail
Notes
Decoder and prompt learner are trained (ImageBind, Vicuna frozen; PandaGPT weights as init). Used for the unsupervised MVTec-AD results and for one-shot transfer to VisA (Section 1; ablation Table 4: 'Unsupervised training and testing are carried out on the MVTec-AD dataset, while the one-shot performance is evaluated on the visa dataset').
Sources
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024) · paper · retrieved 2026-10-15
- AnomalyGPT README (repo now CASIA-LMC-Lab/AnomalyGPT) · readme · retrieved 2026-10-15