GeoStereo: A Unified Stereo Geometry Estimation Framework for Disparity and Surface Normal

Qizhe Wei, Xianda Guo, Shaocong Xu, Hong Li, Runyi Yang, Hao Zhao

GeoStereo

Official implementation of:

GeoStereo: A Unified Stereo Geometry Estimation Framework for Disparity and Surface Normal

GeoStereo is a unified stereo geometry estimation framework that jointly predicts disparity and surface normal by combining stereo matching with diffusion-based geometric priors.

Paper

A Unified Stereo Geometry Estimation Framework for Disparity and Surface Normal

arXiv: https://arxiv.org/abs/2607.24024

Project Links

• Project Page: https://qz-wei.github.io/GeoStereo.github.io/

• Hugging Face: https://huggingface.co/qz-wei/GeoStereo

• Paper: https://arxiv.org/abs/2607.24024

Method Overview

GeoStereo explores the joint estimation of stereo disparity and surface normal within a unified geometric framework.

The framework consists of:

• A stereo matching branch for disparity estimation.
• A diffusion-based normal estimation branch for surface geometry modeling.
• Geometry-aware feature interaction between disparity and normal representations.
• Joint optimization that enables mutual enhancement between stereo correspondence and surface geometry.

Model Zoo

Model	Description
GeoStereo	Full model checkpoint containing stereo and diffusion-based normal estimation modules

Download

Using Hugging Face CLI

hf download qz-wei/GeoStereo