3DMMFN

Introduction

This repository contains the official source code for 3D-MMFN: Multi-level Multimodal Fusion Network for 3D Industrial Image Anomaly Detection.

Requirement

• Ubuntu 18.04

Environment

• Python >= 3.8

Packages

• torch
• torchvision
• numpy
• opencv-python

Other required package(s) can be installed via:

Install pointnet2_ops_lib

pip install "git+git://github.com/erikwijmans/Pointnet2_PyTorch.git#egg=pointnet2_ops&subdirectory=pointnet2_ops_lib"

RUN Model

For RGB Only model:

bash run_RgbOModel.sh

For 3DMMFN model:

bash run_3dmmfn.sh

Data Download and Preprocess

Dataset

The MVTec-3D AD dataset can be downloaded from the Official Website of MVTec-3D AD.

The Eyecandies dataset can be downloaded from the Official Website of Eyecandies.

After downloading, move the dataset to the data folder.

Data Preprocessing

To run the preprocessing for background removal and resizing, run the following command:

python preprocess/preprocessing_mvtec3d.py

Citation

If you find this repository useful for your research, please use the following citation:

@article{asad20253d,
  title={3D-MMFN: Multi-level multimodal fusion network for 3D industrial image anomaly detection},
  author={Asad, Mujtaba and Azeem, Waqar and Malik, Aftab Ahmad and Jiang, He and Ali, Ahmad and Yang, Jie and Liu, Wei},
  journal={Advanced Engineering Informatics},
  volume={65},
  pages={103284},
  year={2025},
  publisher={Elsevier}
}

Acknowledgement

This repo is based on SimpleNet and ShapeGuided, and we thank them for their great work.