CamVid-30K

Summary

This is the CamVid-30K dataset introduced in our paper, "GenXD: Generating Any 3D and 4D Scenes." CamVid-30K is the first open-sourced, large-scale 4D dataset, designed to support various dynamic 3D tasks. It includes videos sourced from VIPSeg, OpenVid-1M, and WebVid-10M, with camera annotations curated using our data curation pipeline.

Project: https://gen-x-d.github.io/

Paper: https://arxiv.org/pdf/2411.02319

Code: https://github.com/HeliosZhao/GenXD

Data curation pipeline.

Data Structure

The zipped dataset is organized in the following structure:

DATA_PATH
└─ camvid-vipseg
   └─ batch_1.zip
   └─ batch_2.zip
└─ camvid-openvid
   └─ batch_1.zip
   └─ batch_2.zip
   └─ ...

After unzipping, each sample contains the images and COLMAP source files:

VIDEO_ID
└─ images
   └─ *.jpg
└─ 0
   └─ cameras.bin
   └─ images.bin
   └─ points3D.bin
   └─ project.ini

Note: Due to the ban on WebVid-10M, we cannot provide the data containing videos from WebVid-10M. If you would like to discuss this subset further, feel free to reach out to Yuyang.

License

Our CamVid-30K dataset is released under CC-BY-4.0. The video samples are collected from publicly available datasets. Users must follow the related licenses VIPSeg, OpenVid-1M, and WebVid-10M to use these video samples.

Citation

@article{zhao2024genxd,
title={GenXD: Generating Any 3D and 4D Scenes},
author={Zhao, Yuyang and Lin, Chung-Ching and Lin, Kevin and Yan, Zhiwen and Li, Linjie and Yang, Zhengyuan and Wang, Jianfeng and Lee, Gim Hee and Wang, Lijuan},
journal={arXiv preprint arXiv:2411.02319},
year={2024}}