YoYo000/MVSNet

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1546 stars · 305 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MVSNet is a deep learning architecture for depth map inference from unstructured multi-view images, and R-MVSNet is its recurrent extension for scalable multi-view stereo reconstruction. This repository contains the official TensorFlow implementation, training/testing scripts, preprocessed datasets, and pretrained models from the ECCV 2018 and CVPR 2019 papers.

Use cases

• infer depth maps from multi-view image sets
• reconstruct 3D point clouds from photos
• train a multi-view stereo network on DTU or BlendedMVS
• evaluate MVS reconstruction on Tanks and Temples or ETH3D
• reproduce ECCV 2018 MVSNet paper results
• run learning-based 3D reconstruction on high-resolution images with R-MVSNet

When to choose

• you need a well-cited, research-grade multi-view stereo depth inference model
• you want pretrained models and preprocessed benchmark datasets ready to use
• you are doing academic work on learning-based 3D reconstruction and need a baseline

When to avoid

• you need a maintained production pipeline - the code requires Python 2.7, CUDA 9.0, and old TensorFlow
• you want a general photogrammetry tool rather than a research codebase
• you lack a GPU or cannot set up the legacy CUDA/TensorFlow environment

Facets

library · maturity maintenance

machine-learning deep-learning computer-vision image-processing computer-vision deep-learning machine-learning python multi-view-stereo depth-estimation 3d-reconstruction tensorflow research-code point-cloud linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="YoYo000/MVSNet")