Reproduce “Exo2EgoDVC: Dense Video Captioning of Egocentric Procedural Activities Using Web Instructional Videos” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Takehiko Ohkawa, Takuma Yagi, Taichi Nishimura, Ryosuke Furuta, Atsushi Hashimoto, Yoshitaka Ushiku, Yoichi Sato
Researchers developed Exo2EgoDVC, a method that enables dense video captioning of egocentric procedural activities by transferring knowledge from abundant exocentric web instructional videos. This approach introduces a new egocentric dataset, EgoYC2, and utilizes view-invariant learning with hand-object features, achieving a dvc eval CIDEr of 59.0 on egocentric cooking videos.
Key insights
- Leveraging abundant exocentric web instructional videos through transfer learning is a practical solution to the data scarcity problem in egocentric DVC.
- Focusing on hand-object interaction through cropping and detailed segmentation masks significantly enhances egocentric video representation for action understanding.
- Gradual domain adaptation, by first aligning visually similar source views and then adapting to the target egocentric view, effectively bridges complex dynamic view gaps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.