We provide fully manually annotated labels on RealGraph dataset, including 2.3M 2D bounding boxes, 760K
2D relationships, 420K 3D bounding boxes and 130K 3D relationships out of 37 object categories and 18 relationship categories in total, each object has a unique identifier across different views and frames. Note that the raw data of RealGraph is 30 FPS, we only annotate semantic labels at 1 FPS. Currently, RealGraph dataset covers 13 dynamic scenes with human activities. Each scene is captured with 8 to 15 cameras with 3 to 20 minutes’ 30Hz HD video, the number of camera is determined by the scale and complexity of the scene.
@inproceedings{lin2023realgraph,
title={RealGraph: A Multiview Dataset for 4D Real-world Context Graph Generation},
author={Lin, Haozhe and Chen, Zequn and Zhang, Jinzhi and Bai, Bing and Wang, Yu and Huang, Ruqi and Fang, Lu},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={3758--3768},
year={2023}
}
We provide fully manually annotated labels on RealGraph dataset, including 2.3M 2D bounding boxes, 760K 2D relationships, 420K 3D bounding boxes and 130K 3D relationships out of 37 object categories and 18 relationship categories in total, each object has a unique identifier across different views and frames. Note that the raw data of RealGraph is 30 FPS, we only annotate semantic labels at 1 FPS. Currently, RealGraph dataset covers 13 dynamic scenes with human activities. Each scene is captured with 8 to 15 cameras with 3 to 20 minutes’ 30Hz HD video, the number of camera is determined by the scale and complexity of the scene.
Paper Code