Researchers Build Massive AI Training Dataset to Perfect Lighting on Faces | HackerNoon
Briefly

The OLAT (One Light at a Time) dataset, utilizing a sophisticated setup of 137 programmable LED lights and 7 frontal-viewing cameras, includes 29,705 unique image sequences with various poses to enhance portrait lighting techniques.
By sourcing HDRI maps from diverse archives such as Polyhaven and HDRMAPS, and incorporating synthetic HDRIs, the study ensured a robust variety in training data, which enhances the method's accuracy during the training phase.
Read at Hackernoon
[
|
]