MVImgNet: A Large-scale Dataset of Multi-view Images

by Xianggang Yu*, Mutian Xu*†, Yidan Zhang*, Haolin Liu*, Chongjie Ye*, Yushuang Wu, Zizheng Yan, Chenming Zhu, Zhangyang Xiong, Tianyou Liang, Guanying Chen, Shuguang Cui, Xiaoguang Han‡ from GAP-Lab.

Introduction

This repository is built for:

MVImgNet: A Large-scale Dataset of Multi-view Images (CVPR2023) [arXiv]

If you find our work useful in your research, please consider citing:

@inproceedings{yu2023mvimgnet,
    title     = {MVImgNet: A Large-scale Dataset of Multi-view Images},
    author    = {Yu, Xianggang and Xu, Mutian and Zhang, Yidan and Liu, Haolin and Ye, Chongjie and Wu, Yushuang and Yan, Zizheng and Liang, Tianyou and Chen, Guanying and Cui, Shuguang and Han, Xiaoguang},
    booktitle = {CVPR},
    year      = {2023}
}
}

MVImgNet

MVImgNet contains 6.5 million frames from 219,188 videos crossing objects from 238 classes. We provide an OneDrive link to download the full data. Please fill out this form to get the download link and password.

We split the full data into 42 zip files, the total size is about 3.4 TB.

Usage

cd path/to/mvimgnet_zip_file
unzip './*.zip'

Folder structure

|-- ROOT
    |-- class_label
        |-- instance_id
            |-- images
            |-- sparse/0
                |-- cameras.bin   # COLMAP reconstructed cameras
                |-- images.bin    # binary data of input images
                |-- points3D.bin  # COLMAP reconstructed sparse point cloud (not dense)

The mapping between class_label and class name can be found in mvimgnet_category.txt.

The images folder contains the multi-view images, and the sparse folder contains the reconstructed camera parameters using COLMAP. It is recommended to use the functions provided by COLMAP to read the binary files under sparse folder. Moreover, the gen_poses function from this repo is recommended to convert the poses for NeRF training.

Script for downloading MVImgNet

We also provide the script, at download_tool.py, for downloading all the content of our dataset. Before using it, please make sure you have filled out our form and get the password.

MVPNet

MVPNet now contains 87,825 point clouds from 180 categories. Please fill out the following form to download MVPNet.

Demo

MVImgNet is also shown by Voxel51 at here, which is publicly demo-able!

License

The data is released under the MVImgNet Terms of Use, and the code is released under the Attribution-NonCommercial 4.0 International License.

Acknowledgement

Thanks to Wei Cheng for the new downloading solution for our dataset.

Thanks to Gege Gao for providing tips on downloading our dataset.

Thanks to Voxel51 for providing the dataset demo.

GAP-LAB-CUHK-SZ/MVImgNet