DAVIAN-Robotics/EgoX

EgoX: Egocentric Video Generation from a Single Exocentric Video

This repository provides model weights of EgoX, a video-to-video generation model that synthesizes egocentric (first-person) videos from a single exocentric (third-person) video. EgoX is built on top of a large-scale video diffusion backbone and enables exo-to-ego viewpoint transformation without requiring multi-view inputs.

For detailed results, implementation details, and demo videos, please refer to our [paper](https://arxiv.org/abs/2512.08269) and [project repository](https://keh0t0.github.io/EgoX/).

Usage

Please refer to the [Quick Start](https://github.com/DAVIAN-Robotics/EgoX?tab=readme-ov-file#-inference) section for instructions on running inference and required preprocessing steps.

Citation

If you find this model or code useful in your research, please cite our paper:

@misc{kang2025egoxegocentricvideogeneration,
  title={EgoX: Egocentric Video Generation from a Single Exocentric Video},
  author={Taewoong Kang and Kinam Kim and Dohyeon Kim and Minho Park and Junha Hyung and Jaegul Choo},
  year={2025},
  eprint={2512.08269},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2512.08269},
}

Acknowledgement

This work builds upon the valuable open-source efforts of 4DNeX and EgoExo4D.

We sincerely appreciate their contributions to the computer vision and robotics communities.