X-LeBench is a novel benchmark and dataset specifically crafted for evaluating tasks on extremely long egocentric video recordings, containing video life logs consist of multiple videos with corresponding timestamps.

Please refer to our paper and project page for details on task definitions and usage guidelines.

📥 To get started:

• Download the video data from Ego4D
• Download simulation and annotation .json files from our HuggingFace dataset page

• 🎫 Life-logging Simulation Pipeline: a novel and customizable pipeline that simulates realistic, ultra-long egocentric video life logs by integrating synthetic daily plans with real-world footages (from Ego4D).
• 🎞️ Dataset: 432 extremely long video life-log simulations of varying lengths.
• 📖 Benchmark Tasks: 4 daily activity related tasks consisting of 8 subtasks.

X-LeBench/
│── assets/                # Images and shared visual assets
│── generation/            # Life-logging simulation pipeline implementation
│── README.md              # Project entry and setup instructions

If you use X-LeBench in your research, please cite our work:

@misc{zhou2025xlebenchbenchmarkextremelylong,
      title={X-LeBench: A Benchmark for Extremely Long Egocentric Video Understanding}, 
      author={Wenqi Zhou and Kai Cao and Hao Zheng and Xinyi Zheng and Miao Liu and Per Ola Kristensson and Walterio Mayol-Cuevas and Fan Zhang and Weizhe Lin and Junxiao Shen},
      year={2025},
      eprint={2501.06835},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2501.06835}, 
}

And please also cite Ego4D for their landmark contributions!

@inproceedings{grauman2022ego4d,
  title={Ego4d: Around the world in 3,000 hours of egocentric video},
  author={Grauman, Kristen and Westbury, Andrew and Byrne, Eugene and Chavis, Zachary and Furnari, Antonino and Girdhar, Rohit and Hamburger, Jackson and Jiang, Hao and Liu, Miao and Liu, Xingyu and others},
  booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
  pages={18995--19012},
  year={2022}
}

This project is licensed under the MIT License. See the LICENSE file for details.