EMatch: A Unified Framework for Event-based Optical Flow and Stereo Matching

Pengjie Zhang, Lin Zhu, Xiao Wang, Lizhi Wang, Hua Huang

This is the official codebase for the paper EMatch: A Unified Framework for Event-based Optical Flow and Stereo Matching (ICCV 2025).

💥 News

• [2025.11.15] ✨ The checkpoints have been organized and published.

• [2025.07.03] 🚀 We released the traning and evaluation code.

• [2025.06.24] 🎉 EMatch is accepted by ICCV 2025.

👀 Introduction

EMatch is a unified framework applicable to both optical flow and stereo matching for event cameras. We reformulate these two tasks as a pixel-wise correspondence matching problem and design a novel unified framework to solve them within a shared representation space. Our unified framework supports both separate training of single-task models for optical flow and disparity, and joint training of a multi-task model — each method achieves optimal performance.

🛠️ Denpendency

You can install anaconda and configure the virtual environment.

The following is a feasible version configuration:

• python = 3.8
• pytorch = 2.0.0
• CUDA = 11.7 (NVIDIA-SMI = 515.76)

You can refer to the following installation steps:

1. Create and activate a virtual environment.

   conda create -n ematch python=3.8
   activate ematch
   

2. Install Pytorch and CUDA.

   conda install pip
   pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1
   

   You can find the installation instructions on the PyTorch official website.

3. Install the necessary python libraries.

   pip install -r requirements.txt
   

🤖️ Checkpoints

You can download all checkpoints here: Google Drive / BaiduDisk

📁 Datasets

Please download the DSEC and MVSEC datasets.

Details about the datasets can be found in the data/README.md file.

📈 Training and Evaluation

All training and evaluation scripts are provided in the /scripts directory.

For example, you can run the .sh files directly from the command line to obtain results of EMatch:

sh ./scripts/test/unified/dsec_ematch_flow.py
sh ./scripts/test/unified/dsec_ematch_disparity.py
sh ./scripts/test/unified/mvsec_ematch_flow.py
sh ./scripts/test/unified/mvsec_ematch_disparity.py

You can run the following .sh files to train EMatch:

sh ./scripts/train/unified/dsec_ematch_stage1.sh
sh ./scripts/train/unified/dsec_ematch_stage2.sh
sh ./scripts/train/unified/mvsec_ematch.sh

More scripts are available in the /scripts directory. If you wish to conduct further experiments, you may customize the provided .sh files to fit your specific requirements.

💌 Acknowledgement

We would like to thank the following projects for their contributions to this work: DCEI, TMA, IDNet, E-RAFT, se-cff, unimatch.