[IEEE SMC Accepted] Multi-View Reconstruction with Global Context for 3D Anomaly Detection.
by Yihan Sun*,Yuqi Cheng*, Yunkang Cao*, Yuxin Zhang, Weiming Shen,
3D anomaly detection is critical in industrial quality inspection. While existing methods achieve notable progress, their performance degrades in high-precision 3D anomaly detection due to insufficient global information. To address this, we propose Multi-View Reconstruction (MVR), a method that losslessly converts high-resolution point clouds into multi-view images and employs a reconstruction-based anomaly detection framework to enhance global information learning. Extensive experiments demonstrate the effectiveness of MVR, achieving 89.6% object-wise AU-ROC and 95.7% point-wise AU-ROC on the Real3D-AD benchmark.
To set up the MVR environment, follow one of the methods below:
- Clone this repo:
git clone https://github.com/hustSYH/MVR.git && cd MVR
- Construct the experimental environment, follow these steps:
conda env create -f environment.yml conda activate MVR
You can choose to download original datasets and process them according to Preprocess. You are also welcome to directly download our processed datasets. All datasets need to be placed in your DATA_ROOT.
| Dataset | Google Drive | Baidu Drive | Note |
|---|---|---|---|
| MVTec3D | [Google Drive] | Baidu Drive | Original |
| Real3D | [Google Drive] | Baidu Drive | Original |
| MVTec3D-multiview | [Google Drive] | Baidu Drive | Processed |
| Real3D-multiview | [Google Drive] | Baidu Drive | Processed |
To enhance feature discriminative capability, the rendered images are fixed with a spatial resolution of 672 × 672 by default. After generating multi-view images at high resolution, they are resized to 224×224 as the input images.
We remove the background and project point clouds to multi_view images.
sh process_MVTec3D.sh
We first convert Real3D to Depth data .tiff file, and then project point clouds to multi_view images. You also can get it from BaidDu Drive
sh process_Real3D.sh
python MVR_mvtec3d.py
python MVR_real3d.py
Our work is largely inspired by the following projects. Thanks for their admiring contribution.
If you find this project helpful for your research, please consider citing the following BibTeX entry.
@inproceedings{MVR,
title={Multi-View Reconstruction with Global Context for 3D Anomaly Detection},
author={Yihan, Sun and Yuqi, Cheng and Yunkang, Cao and YUxin, Zhang and Weiming, Shen},
booktitle={},
year={2025}
}