Meta AI Releases Sapiens2: A High-Resolution Human-Centric Vision Model for Pose, Segmentation, Normals, Pointmap, and Albedo

Meta AI released Sapiens2, a second-generation human-centric vision foundation model for dense person understanding tasks.

It was trained on a curated 1B-image dataset and uses both masked image reconstruction and contrastive pretraining.

The model improves pose estimation, segmentation, surface normals, pointmaps, and albedo prediction at 1K and 4K resolution.

Meta says the broader detail and generalization could benefit AR/VR, motion capture, digital humans, and photo-realistic editing.