SAIL-VOS 3D is a synthetic video dataset with frame-by-frame 3D mesh annotations for objects, ground-truth depth for scenes and 2D annotations in the form of bounding boxes, instance-level semantic segmentation and amodal segmentation. It contains instance-level 3D annotation that captures dynamic scenes with diverse scenarios.

Dataset Statistics

SAIL-VOS 3D (S3D) dataset contains in total 237,611 video frames, 3,460,213 instances being annotated with 3D mesh and 2D semantic/amodal segmentation. There are in total 178 object categories in this dataset. In addition to the training and validation sets, we retain a test-dev set and a test-challenge set for future use.

Download Dataset

Please contact us for the link to download the data. See the dataset [README].
Before downloading the data you must agree to the following terms:
1. You will use the data only for non-commercial research and educational purposes. Commercial use is prohibited.
2. You will NOT distribute the data.
3. You buy Grand Theft Auto V.

Publications

Please cite the following papers if you find the dataset useful.

@inproceedings{HuCVPR2021,
  author = {Y.-T. Hu and J. Wang and R.~A. Yeh and A.~G. Schwing},
  title = { {SAIL-VOS 3D: A Synthetic Dataset and Baselines for Object Detection and 3D Mesh Reconstruction from Video Data} },
  booktitle = {Proc. CVPR},
  year = {2021},
}

@inproceedings{HuCVPR2019,
  author = {Y.-T. Hu and H.-S. Chen and K. Hui and J.-B. Huang and A.~G. Schwing},
  title = { {SAIL-VOS: Semantic Amodal Instance Level Video Object Segmentation -- A Synthetic Dataset and Baselines} },
  booktitle = {Proc. CVPR},
  year = {2019},
}