EgoDynamic4D Dataset from AAAI 2026 paper: Understanding Dynamic Scenes in Egocentric 4D Point Clouds
🚀 EgoDynamic4D dataset is coming soon.
This repository hosts the official release of EgoDynamic4D, a large-scale egocentric 4D dynamic scene understanding benchmark introduced in our AAAI 2026 paper:
Understanding Dynamic Scenes in Egocentric 4D Point Clouds
About the Dataset
EgoDynamic4D is a question answering (QA) benchmark designed for fine-grained spatio-temporal reasoning in egocentric dynamic scenes.
The dataset includes:
- Egocentric RGB-D videos
- Camera poses
- Globally unique instance masks
- 4D bounding boxes over time
- Large-scale QA annotations for dynamic reasoning tasks
The dataset is constructed based on existing egocentric 4D resources, building upon ADT and THUD++. Our main contribution lies in the large-scale, task-driven QA annotations, which enable fine-grained spatio-temporal reasoning in dynamic egocentric scenes.
See also on Github