SoroushMehraban/3D-Pain

3D Pain

3D pain is the dataset proposed in Pain in 3D: Generating Controllable Synthetic Faces for Automated Pain Assessment. It is made of 2,500 synthetic faces each having 10 different variants of pain, rendered from three different viewpoints with colorized heatmaps.

Paper arXiv: https://arxiv.org/abs/2509.16727

Dataset file structure:

.
├── heatmaps/  # 25k colorized heatmaps
│   ├── README.txt
│   ├── heatmaps_00000.tar.gz
│   ├── heatmaps_00001.tar.gz
│   └── ....
├── textured_meshes/ # 2.5k textured meshes
│   ├── README.txt
│   ├── meshes_00000.tar.gz
│   └── meshes_00001.tar.gz
├── images/  # 82.5k images (27.5k images in 3 different views, among them 2.5k are neutral and 25k are having pain)
│   ├── README.txt
│   ├── images_00000.tar.gz
│   ├── images_00001.tar.gz
│   └── ...
└── annotations/ # 25k pain annotations
    ├── README.txt
    ├── annotations_00000.tar.gz
    ├── annotations_00001.tar.gz
    └── ...

Inside each folder there is README.txt that gives the instruction to extract the files.

If you find our work useful for your project, please consider citing the paper:

@article{lin2025pain,
  title={Pain in 3D: Generating Controllable Synthetic Faces for Automated Pain Assessment},
  author={Lin, Xin Lei and Mehraban, Soroush and Moturu, Abhishek and Taati, Babak},
  journal={arXiv preprint arXiv:2509.16727},
  year={2025}
}