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

Crestereo

CREStereo is an open-source MegEngine implementation of a stereo-matching model that takes left and right images and produces a disparity map. Its repository provides a pretrained model and inference code, as well as training code and stereo datasets.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Crestereo — 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 CREStereo when a project needs disparity estimation from paired stereo images and the team can work with the repository’s Python and MegEngine environment.

Who it is for

  • Computer-vision developers building stereo-vision or depth-related perception workflows.
  • Researchers who want to run or train a stereo-matching model using MegEngine.
  • Teams evaluating disparity estimation on paired camera images.

Documented uses

  • Generate a disparity map from a left- and right-camera image pair.
  • Use a pretrained MegEngine model with the provided inference script.
  • Configure and run the training code for a stereo-matching workflow.
  • Inspect disparity data from the proposed or listed public stereo datasets.
Before you choose

Requirements & limits

The implementation is publicly available on GitHub under the Apache-2.0 license. The README documents a Docker example as well as a CUDA-based Python environment.

  • The model requires paired left and right images rather than a single image.
  • The documented stack specifies CUDA 10.1, Python 3.6.9 and MegEngine 1.8.2, plus OpenCV, NumPy, Pillow and TensorBoardX.
  • The repository’s proposed training dataset is approximately 400 GB; other public stereo datasets are also listed.
  • Inference requires downloading the pretrained MegEngine model linked from the README.

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 inference on a stereo image pair

Generate a disparity-map output from left and right images using the repository’s pretrained model and test script.

Before you start

  • The CREStereo repository and its documented Python dependencies, including MegEngine.
  • A downloaded pretrained MegEngine model.
  • A left image and a corresponding right image.

Steps

  1. Download the pretrained MegEngine model linked in the repository README.

  2. Use the repository’s test.py command with the model path, left and right image paths, size setting and desired output path.

  3. Read the generated disparity map at the specified output path.

Expected outcome

A disparity map is written to the output path specified for inference.

Guide limitations

  • The README specifies CUDA 10.1 and Python 3.6.9, with MegEngine 1.8.2 and additional dependencies.
  • The guide reflects documented commands and does not establish compatibility with other environments.
Documentation 1 ↗
Source-based practical guide

Prepare the proposed CREStereo training dataset

Download and extract the repository’s proposed dataset using its shell script.

Before you start

  • A local copy of the repository, including dataset_download.sh.
  • Storage for the approximately 400 GB dataset.

Steps

  1. From the repository directory, run the command below.

    Documented command
    sh dataset_download.sh

  2. Use the downloaded and extracted dataset under ./stereo_trainset/crestereo.

  3. For disparity PNG files, load them as uint16 with OpenCV and divide the values by 32 to obtain float disparity values, as shown in the README.

Expected outcome

The proposed dataset is downloaded and extracted in the documented directory, with disparity values interpretable using the documented scale.

Guide limitations

  • The README describes this dataset as approximately 400 GB.
  • This recipe covers dataset preparation, not launching or completing model training.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
CrestereoGenerate a disparity map from a left- and right-camera image pair.Docker, LinuxThe implementation is publicly available on GitHub under the Apache-2.0 license. The README documents a Docker example as well as a CUDA-based Python environment.
Pulse Of MotionEstimate a video's physical frame rate from its visual dynamics.LinuxThe repository is public. Its README describes a released checkpoint that downloads from Hugging Face on first inference, with a manual download option. The strongest internal model is not fully released; the README directs interested users to contact the project about access. No license information is provided in the repository evidence.
Neural Image AssessmentTrain the model using AVA images and rating annotations.Cloud, Linux, macOS, Windows
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
megvii-research/CREStereo · Official homepage title · retained in its source language
How we review sourcesSuggest a correction