OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Unidepth

UniDepth is a Python computer-vision model project that estimates metric depth from a single RGB image and can also predict camera intrinsics and 3D points in camera coordinates.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Unidepth — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use UniDepth when a workflow needs metric depth and related camera-geometry predictions from a single RGB image, and you can run the documented Python model environment.

Who it is for

  • Computer-vision developers building monocular depth-estimation pipelines.
  • Researchers and practitioners using single-image depth and camera geometry in 3D reconstruction or spatial scene analysis.

Documented uses

  • Estimate metric scene depth from one RGB image.
  • Generate camera-coordinate 3D points and predict camera intrinsics for downstream vision workflows.
  • Add a pretrained monocular depth model to a Python computer-vision pipeline.
Before you choose

Requirements & limits

  • The documented inference example uses a pretrained UniDepthV1 model loaded from Hugging Face.
  • The README lists Linux, Python 3.10+, and CUDA 11.8+; it describes these as potential rather than absolute requirements.
  • The package can produce depth, camera-coordinate points, and intrinsics from RGB input; ground-truth intrinsics can also be supplied.
  • The repository documents a demo script for checking installation and notes possible CUDA/xFormers environment issues.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Run metric-depth inference on an RGB image

Use a pretrained UniDepthV1 model to obtain depth and camera-geometry predictions from an RGB image.

Before you start

  • Linux, Python 3.10+, and CUDA 11.8+ are listed by the README; the project notes these may not be hard requirements.
  • Install UniDepth and its dependencies in a Python virtual environment or Conda environment.
  • A pretrained UniDepthV1 model and an RGB image.

Steps

  1. Install UniDepth and dependencies using the repository’s documented pip instructions, or follow its Conda environment alternative.

  2. Load a pretrained model with UniDepthV1.from_pretrained, choosing one of the model identifiers shown in the README.

  3. Convert the RGB image to a channel-first tensor and pass it to model.infer.

  4. Read the returned depth, points, and intrinsics predictions; optionally provide ground-truth camera intrinsics as described in the README.

Expected outcome

Inference returns metric depth, 3D points in camera coordinates, and predicted camera intrinsics.

Guide limitations

  • CUDA compilation and runtime versions should match; the README describes potential xFormers/CUDA issues that may slow inference.
  • The project notes that environment requirements may not be hard requirements and that differences are not tested.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
lpiccinelli-eth/UniDepth · Official homepage title · retained in its source language
How we review sourcesSuggest a correction