---
title: "RAFT: Recurrent All-Pairs Field Transforms for Optical Flow (Extended Abstract)"
authors:
  - "Zachary Teed"
  - "Jia Deng"
doi: "10.24963/ijcai.2021/662"
url: "https://ijcai.org/proceedings/2021/662"
pdf: "https://ijcai.org/proceedings/2021/0662.pdf"
venue: "Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence"
year: 2021
pages: "4839-4843"
track: "Sister Conferences Best Papers"
keywords:
  - "Computer Vision: 2D and 3D Computer Vision"
  - "Computer Vision: Motion and Tracking"
---

# RAFT: Recurrent All-Pairs Field Transforms for Optical Flow (Extended Abstract)

Zachary Teed, Jia Deng

*Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence*, pages 4839-4843. DOI: [10.24963/ijcai.2021/662](https://doi.org/10.24963/ijcai.2021/662)

## Abstract

We introduce Recurrent All-Pairs Field Transforms (RAFT), a new deep network architecture for optical flow. RAFT extracts per-pixel features, builds multi-scale 4D correlation volumes for all pairs of pixels, and iteratively updates a flow field through a recurrent unit that performs lookups on the correlation volumes. RAFT achieves state-of-the-art performance on the KITTI and Sintel datasets. In addition, RAFT has strong cross-dataset generalization as well as high efficiency in inference time, training speed, and parameter count.