Skip to content

Goliath-4 dataset

For four subjects, we provide paired scans of faces and heads with lighting information, as well as scans of bodies in both clothed and minimally-clothed modes.

4

relightable head dome sessions

4

relightable hands dome sessions

4

fully clothed sessions

4

minimally-clothed sessions

Abstract

Meta Reality Labs Research introduces Goliath-4, a multi-modal dataset collecting relightable high-resolution heads and hands, together with complete full-body scans of four subjects wearing both casual and minimal clothing. Additionally, phone camera images of these four modalities are provided.

Researchers in different areas of body modeling will benefit from these raw images, the provided assets computed from them (keypoints, semantic segmentation, 3D reconstruction and registrations), as well as from the code and models from state-of-the-art publications already using this data. Goliath-4 creates the opportunity to link research done on low resolution inputs with the highest quality avatars.

Models and code included in release

Relightable head decoder

We provide a model of a personalized relightable head avatar based on Gaussian splatting.

Body decoders

We provide mesh-based body decoders with neural textures.