Skip to content

Depth Pro

Fast monocular depth maps from a single image

Visit official site

What Depth Pro is

The model is described as a research project for monocular depth estimation, producing accurate depth maps with a multi-scale vision transformer for dense prediction.

What you can do with it

  • Generate a high-resolution depth map from one image
  • Estimate focal length within a single image
  • Apply boundary-focused evaluation metrics

Getting started

  • Set up a virtual environment such as miniconda
  • Download pretrained models with get_pretrained_models.sh
  • Run depth-pro-run or call the model in Python

What to watch out for

  • Training mixes real and synthetic datasets
  • A 2.25 megapixel map is stated to take 0.3 seconds
  • Named uses include AR and autonomous driving

Pros & cons

✓ What we like

  • A dedicated metric for depth map boundary accuracy
  • Focal length estimation from a single image

! What to watch out for

  • Setup asks for a virtual environment and pretrained models
  • Inference runs through a command line tool or Python script

FAQ

How fast is it stated to be?

The material states 0.3 seconds for a 2.25 megapixel depth map on a standard GPU.

How is it run?

The steps name the depth-pro-run command and a Python script.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review