I hope you are doing well. I have been studying your FaceOLAT dataset and the accompanying code repository, as well as your paper on 3D portrait relighting. The dataset and pipeline are extremely helpful for my research.
While working with the relighting pipeline, I encountered some questions regarding the construction of the voronoi_indices.exr index map and was hoping you could kindly clarify a few details:
- Index map construction
Could you please confirm how the voronoi_indices.exr is generated?
Specifically, is it constructed via a spherical Voronoi partition based on the light directions (e.g., assigning each direction on the sphere to the nearest light)?
2. Projection / parameterization
What spherical parameterization is used when rasterizing the index map?
For example, is it a standard latitude-longitude (equirectangular) projection?
3. Coordinate system alignment
I am particularly interested in how the environment map is aligned with the light stage:
Where is the reference direction (e.g., φ = 0 / “prime meridian”) in the index map?
Which direction corresponds to the subject’s frontal direction in the dataset coordinate system?
What is the convention for horizontal rotation (clockwise vs counterclockwise)?
4. Consistency across subjects
Am I correct in understanding that the same index map and light ordering are shared across all subjects in the dataset?
Understanding these details would greatly help ensure correct alignment when using custom environment maps for relighting.
Thank you very much for your time and for making this dataset publicly available. I truly appreciate your work.
I hope you are doing well. I have been studying your FaceOLAT dataset and the accompanying code repository, as well as your paper on 3D portrait relighting. The dataset and pipeline are extremely helpful for my research.
While working with the relighting pipeline, I encountered some questions regarding the construction of the voronoi_indices.exr index map and was hoping you could kindly clarify a few details:
Could you please confirm how the voronoi_indices.exr is generated?
Specifically, is it constructed via a spherical Voronoi partition based on the light directions (e.g., assigning each direction on the sphere to the nearest light)?
2. Projection / parameterization
What spherical parameterization is used when rasterizing the index map?
For example, is it a standard latitude-longitude (equirectangular) projection?
3. Coordinate system alignment
I am particularly interested in how the environment map is aligned with the light stage:
Where is the reference direction (e.g., φ = 0 / “prime meridian”) in the index map?
Which direction corresponds to the subject’s frontal direction in the dataset coordinate system?
What is the convention for horizontal rotation (clockwise vs counterclockwise)?
4. Consistency across subjects
Am I correct in understanding that the same index map and light ordering are shared across all subjects in the dataset?
Understanding these details would greatly help ensure correct alignment when using custom environment maps for relighting.
Thank you very much for your time and for making this dataset publicly available. I truly appreciate your work.