Apple debuts Depth Pro 2 for video
Apple released Depth Pro 2, a foundation model family delivering sharp, temporally consistent metric monocular depth estimation across static images and video feeds. The architecture pairs spatial and temporal attention with zero-shot focal length estimation to produce metric depth without pre-calibrated camera sensors.
Depth Pro 2 eliminates the traditional trade-off between edge sharpness and temporal stability in monocular depth, making real-time spatial computing practical on standard cameras without LiDAR.
- –Joint spatial-temporal attention prevents frame-to-frame depth flickering while preserving fine-grained boundaries around hair, fur, and thin structures.
- –Predicts true metric depth in meters without pre-calibrated intrinsics, using an integrated zero-shot field-of-view estimator.
- –Multi-tier model family ranges from a 24M-parameter Small variant optimized for Core ML and edge devices to an 872M High-capacity model.
- –Core ML sliding-window support enables real-time stream processing on Apple Silicon, unlocking live AR and video editing workflows.
DISCOVERED
6h ago
2026-11-22
PUBLISHED
6h ago
2026-11-22
RELEVANCE
AUTHOR
AI Search