Apple’s Machine Learning Research team today published Depth Pro 2, a system that performs dense metric depth estimation on images and video without downscaling.
According to Apple, the model produces exact depth information, maintains crisp edges, and preserves consistent frames across video, beating earlier approaches on every front. The release encompasses several models that together set a new benchmark for balancing precision, detail, smoothness, and processing time.
Apple has released the code and trained weights for Depth Pro 2 on its machine learning research portal.
Why it matters
Depth Pro 2 is the successor to the Depth Pro model Apple shipped in October 2024, which created detailed depth maps from single images in 0.3 seconds. In December 2025, Apple also published SHARP, which can generate photorealistic views from one image by predicting 3D Gaussian parameters in less than a second. With the open release of Depth Pro 2, Apple is again making its computer vision advances available to researchers and developers.
