Kevin Qinghong Lin, Linjie Li, Difei Gao, Qinchen Wu, Mingyi Yan, Zhengyuan Yang, Lijuan Wang, Mike Zheng Shou
- [2024.6.17] We release the arXiv paper.
TL;DR: A Multi-modal Benchmark for Visual-centric GUI Automation from Instructional Videos.
Visual-centric softwares and tasks: VideoGUI focuses on professional and novel software like PR and AE for video editing, or Stable Diffusion and Runway for visual creation. Besides, the task query emphasizes visual preview rather than textual instructions.
Instructional videos with human demonstration: We source novel tasks from high-quality instructional videos, with annotators replicating these to reproduce effects.
Hierarchical planning and actions: We provide detailed annotations with planning procedures and recorded actions for hierarchical evaluation.