| --- |
| task_categories: |
| - question-answering |
| language: |
| - en |
| --- |
| # EgoDynamic4D Dataset from AAAI 2026 paper: [Understanding Dynamic Scenes in Egocentric 4D Point Clouds](https://arxiv.org/abs/2508.07251) |
|
|
| 🚀 **EgoDynamic4D QA dataset has been released.** |
|
|
| This repository hosts the official implementation 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](https://arxiv.org/abs/2306.06362) and [THUD++](https://arxiv.org/abs/2412.08096). Our main contribution lies in the **large-scale, task-driven QA annotations**, which enable fine-grained spatio-temporal reasoning in dynamic egocentric scenes. |
|
|
| **Please find instructions on [Github](https://github.com/Dancing-Github/EgoDynamic4D) for usages.** |
