[TMM 2025] Learning Multi-view Anomaly Detection with Efficient Adaptive Selection

Haoyang He¹, Jiangning Zhang^1†, Guanzhong Tian¹, Chengjie Wang², Lei Xie^1†

¹College of Control Science and Engineering, Zhejiang University, ²Youtu Lab, Tencent,

[Paper]

Our MVAD is based on ADer.

Congratulations! Our MVAD has been accepted at the IEEE Transactions on Multimedia!

📜 Multi-class Results on Real-IAD Multi-View Setting

Subscripts S, I, and R represent rsample-level, image-level, and pixel-level, respectively.

MVAD Results

Dataset	mAU-ROCS	mAPS	mF1-maxS	mAU-ROCI	mAPI	mF1-maxI	mAU-ROCP	mAPP	mF1-maxP	mAU-PROP	Download
Real-IAD	90.2	95.3	90.1	86.6	84.8	77.2	97.9	30.3	36.8	91.2	log & weight

🛠️ Getting Started

Installation

• Prepare general experimental environment

  pip3 install timm==0.8.15dev0 mmselfsup pandas transformers openpyxl imgaug numba numpy tensorboard accimage Ninja
  pip3 install --upgrade protobuf==3.20.1 scikit-image faiss-gpu
  pip3 install geomloss FrEIA adeval fvcore==0.1.5.post20221221
  pip3 install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118
  (or) conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=11.8 -c pytorch -c nvidia

Dataset Preparation

Multi-view Real-IAD Dataset

• Download and extract Real-IAD into data/realiad.

Train (Multi-view Anomaly Detection under Multi-class Unsupervised AD Setting)

• Check data and model settings for the config file configs/mvad/mvad_realiad.py
• Train with single GPU example:

CUDA_VISIBLE_DEVICES=0 python run.py -c configs/mvad/mvad_realiad.py -m train

• Modify trainer.resume_dir to resume training.

Test

• The training log of MVAD can be find at log and the weights at model.
• Modify trainer.resume_dir or model.kwargs['checkpoint_path']
• Test with single GPU example:

CUDA_VISIBLE_DEVICES=0 python run.py -c configs/mvad/mvad_realiad.py -m test model.kwargs.checkpoint_path=log/mvad.pth

Citation

If you find this code useful, don't forget to star the repo and cite the paper:

@article{he2024learning,
  title={Learning Multi-view Anomaly Detection},
  author={He, Haoyang and Zhang, Jiangning and Tian, Guanzhong and Wang, Chengjie and Xie, Lei},
  journal={arXiv preprint arXiv:2407.11935},
  year={2024}
}