Depth Pro

Depth Pro was unveiled by Apple's machine learning research team in October 2024 (paper arXiv 2410.02073, public announcement around October 7-10). It is a foundation model for zero-shot metric monocular depth estimation: given a single image, it synthesizes a 2.25-megapixel depth map with absolute scale in about 0.3 seconds on a standard GPU, without requiring camera intrinsics or other metadata.

The design combines an efficient multi-scale vision transformer with a training protocol mixing real and synthetic data for both metric accuracy and sharp boundary tracing, plus state-of-the-art focal-length estimation from the image alone. Apple released code and weights at github.com/apple/ml-depth-pro. The release uses Apple's sample-code licensing, so verify the terms on the repo before commercial use.