DepthPro

Depth Pro is a foundation model for zero-shot metric monocular depth estimation. It synthesizes high-resolution depth maps with sharpness and high-frequency detail. The predictions are metric with absolute scale and do not require camera intrinsics. The model is fast, producing a 2.25-megapixel depth map in 0.3 seconds on a standard GPU.

Key Strengths

• Zero-shot metric depth estimation from a single image.
• High-resolution output with sharp boundary tracing.
• No dependency on metadata such as camera intrinsics.
• State-of-the-art focal length estimation from a single image.
• Efficient multi-scale vision transformer architecture.

Best For

Applications requiring precise, metric depth maps with fine detail—such as 3D reconstruction, augmented reality, and robotics—where inference speed and camera-agnostic operation are critical.

Performance

Depth Pro produces a 2.25-megapixel depth map in 0.3 seconds on a standard GPU. The checkpoint is a re-trained implementation; its performance is close to, but does not exactly match, the model reported in the paper.

Reference

Introduced in Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (Bochkovskii et al., 2024).