BASKET-Multiview Dataset

The next 4D Human-Centric Benchmark

We introduce the BASKET (BAsketball Synthetic benchmarK for Enhanced Telepresence)-Multiview Dataset, a synthetic collection of scenarios representing common basketball plays generated using Unreal Engine 5 and EasySynth. For each scene, we provide comprehensive annotations that include calibrated cameras parameters, animations, RGB images, segmentation masks, depth maps, surface normal images, SMPL meshes and animations. All scenes are rendered at 1080p/4K and 30 fps.

Dataset Overview

BASKET provides multi-modal annotations including RGB images, depth maps, segmentation masks, surface normals, and SMPL meshes for dynamic basketball scenes.

Why BASKET-Multiview?

The lack of comprehensive dynamic 3D datasets focusing on human-centric events has been a significant bottleneck in advancing real-time 3D reconstruction and telepresence technologies. Existing datasets either focus on static scenes, lack multi-view coverage, or provide insufficient ground truth annotations for rigorous evaluation.

BASKET addresses these limitations by providing a high-quality synthetic benchmark specifically designed for evaluating dynamic human-centric 3D reconstruction methods. Our dataset features realistic basketball gameplay scenarios with comprehensive multi-modal annotations, making it an ideal testbed for algorithms targeting live sports streaming and immersive viewing experiences.

We envision BASKET becoming the standard benchmark for evaluating dynamic 3D reconstruction methods, particularly those focusing on human activities and real-time performance requirements.

Dataset Statistics

89
Camera Views
7
Full plays
2
Scenarios
9K+
Total Frames
1080p/4K
Resolution

What's Included

Visual Data

  • RGB Images: High-quality images from all cameras
  • Depth Maps: Accurate depth information for each frame
  • Segmentation: Per-player and object masks
  • Normals: Geometric surface information

3D Annotations

  • Camera Parameters: Full calibration data for all views
  • Animations: Full motion capture data
  • Artefacts: Upcoming
  • SMPL Meshes: Upcoming

Download Dataset

To request access to the dataset, please complete the form below using an official institutional or company email address.
If you've already registered, please use this link to browse/download.

Please use your institutional or company email address

Thank you for your interest in our work!

You can now download the full dataset or browse/partially download using the following credentials.

Username: basket_multiview
Password: &]03~09lFk+l

Environment Assets

The BASKET-Multiview dataset was created using high-quality 3D assets purchased from various marketplace vendors. While we cannot redistribute these assets directly due to licensing restrictions, researchers interested in recreating or extending our environment can obtain them from their original sources:

Please note that purchasing these assets separately may be required to fully recreate the environment shown in our dataset. We recommend checking the individual license terms from each vendor for your intended use case.

BibTeX

@article{10.1145/3731214,
  author = {Huang, Junkai and Subhajyoti Mallick, Saswat and Amat, Alejandro and Ruiz Olle, Marc and Mosella-Montoro, Albert and Kerbl, Bernhard and Vicente Carrasco, Francisco and De la Torre, Fernando},
  title = {Echoes of the Coliseum: Towards 3D Live streaming of Sports Events},
  year = {2025},
  issue_date = {August 2025},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  volume = {44},
  number = {4},
  issn = {0730-0301},
  url = {https://doi.org/10.1145/3731214},
  doi = {10.1145/3731214},
  journal = {ACM Trans. Graph.},
  month = jul,
  articleno = {46},
  numpages = {17},
  keywords = {sports events, human-centric events, gaussian splatting, real-time 3D reconstruction}
}