[CVPR 2026] Reproduce for Ego2Web: A Web Agent Benchmark Grounded in Egocentric Videos

Authors

Shoubin Yu^1,2 · Lei Shu¹ · Antoine Yang¹ · Yao Fu¹ · Srinivas Sunkara¹ · Maria Wang¹ · Jindong Chen¹ · Mohit Bansal² · Boqing Gong¹

¹ Google DeepMind ² University of North Carolina at Chapel Hill

📖 Overview

Ego2Web is a benchmark for evaluating web agents grounded in egocentric video understanding. It connects real-world human activities with web-based tasks, enabling research at the intersection of embodied perception and web interaction.

📌 Citation

If you find this work useful, please cite:

@article{yu2026ego2web,
  title={Ego2Web: Benchmarking Web Agents with Egocentric Video Grounding},
  author={Yu, Shoubin and Shu, Lei and Yang, Antoine and Fu, Yao and 
          Sunkara, Srinivas and Wang, Maria and Chen, Jindong and 
          Bansal, Mohit and Gong, Boqing},
  journal={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026}
}