---
title: "NeuFlow v2: High-Efficiency Optical Flow Estimation on Edge Devices"
canonical_url: "https://www.modelscope.cn/papers/43893"
md_url: "https://www.modelscope.cn/papers/43893.md"
arxiv_id: 2408.10161
published: 2024-08-19
last_updated: 2024-08-19
authors:
  - "Zhiyong Zhang"
  - "Aniket Gupta"
  - "Huaizu Jiang"
  - "Hanumant Singh"
model_name: "NeuFlow v2"
model_developer: "美国东北大学"
domain:
  - "计算机视觉"
  - "深度学习"
type:
  - "计算机视觉"
  - "深度学习"
  - "Computer Vision and Pattern Recognition (cs.CV)"
arxiv_url: "https://arxiv.org/abs/2408.10161"
pdf_url: "https://arxiv.org/pdf/2408.10161.pdf"
code_link: "https://github.com/neufieldrobotics/NeuFlow_v2"
---

# NeuFlow v2: High-Efficiency Optical Flow Estimation on Edge Devices

> Real-time high-accuracy optical flow estimation is crucial for various real-world applications. While recent learning-based optical flow methods have achieved high accuracy, they often come with significant computational costs. In this paper, we propose a…

「NeuFlow v2: High-Efficiency Optical Flow Estimation on Edge Devices」是 ModelScope 魔搭社区收录的论文，arXiv 2408.10161，作者为 Zhiyong Zhang, Aniket Gupta, Huaizu Jiang et al.，发表于 2024-08-19，属于 计算机视觉、深度学习 领域。

- **ArXiv**: 2408.10161
- **Published**: 2024-08-19
- **Authors**: Zhiyong Zhang, Aniket Gupta, Huaizu Jiang, Hanumant Singh
- **Model**: NeuFlow v2
- **Developer**: 美国东北大学
- **Domain**: 计算机视觉, 深度学习
- **ArXiv URL**: https://arxiv.org/abs/2408.10161
- **PDF**: https://arxiv.org/pdf/2408.10161.pdf
- **Code**: https://github.com/neufieldrobotics/NeuFlow_v2

Source: https://www.modelscope.cn/papers/43893

---

> NeuFlow v2：在边缘设备上实现高效光流估计的新方法

## 摘要

本文提出了一种高效的光流估计算法NeuFlow v2，旨在在边缘设备上实现高精度和低计算需求的平衡。基于NeuFlow v1，本文引入了一个更轻量级的骨干网络和一个快速的迭代细化模块。这些新组件在保持接近最先进精度的同时，大大降低了计算需求。与其他最先进的方法相比，NeuFlow v2在合成和真实数据上实现了10倍到70倍的速度提升，并且在Jetson Orin Nano设备上以超过20帧每秒的速度运行512x384分辨率的图像。本文的主要贡献包括：1）提出了一个简单的CNN骨干网络，用于从多尺度图像中提取低级特征；2）设计了一个轻量级的迭代细化模块，能够输出隐藏状态和解码后的细化光流。实验结果表明，NeuFlow v2在保持高精度的同时，大大提高了计算效率，特别适用于边缘计算平台。

## Abstract

Real-time high-accuracy optical flow estimation is crucial for various real-world applications. While recent learning-based optical flow methods have achieved high accuracy, they often come with significant computational costs. In this paper, we propose a highly efficient optical flow method that balances high accuracy with reduced computational demands. Building upon NeuFlow v1, we introduce new components including a much more light-weight backbone and a fast refinement module. Both these modules help in keeping the computational demands light while providing close to state of the art accuracy. Compares to other state of the art methods, our model achieves a 10x-70x speedup while maintaining comparable performance on both synthetic and real-world data. It is capable of running at over 20 FPS on 512x384 resolution images on a Jetson Orin Nano. The full training and evaluation code is available at this https URL.