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.
Keywords
RaftComputer sciencePixelAlgorithmGeneralizationField (mathematics)Artificial intelligenceMathematicsPhysics
Affiliated Institutions
Related Publications
Lipid Domain Structure of the Plasma Membrane Revealed by Patching of Membrane Components
Lateral assemblies of glycolipids and cholesterol, “rafts,” have been implicated to play a role in cellular processes like membrane sorting, signal transduction, and cell adhesi...
Association of Sterol- and Glycosylphosphatidylinositol-linked Proteins with Drosophila Raft Lipid Microdomains
In vertebrates, the formation of raft lipid microdomains plays an important part in both polarized protein sorting and signal transduction. To establish a system in which raft-d...
Sphingolipid–Cholesterol Rafts Diffuse as Small Entities in the Plasma Membrane of Mammalian Cells
To probe the dynamics and size of lipid rafts in the membrane of living cells, the local diffusion of single membrane proteins was measured. A laser trap was used to confine the...
Publication Info
- Year
- 2020
- Type
- book-chapter
- Pages
- 402-419
- Citations
- 2044
- Access
- Closed
External Links
Social Impact
Altmetric
PlumX Metrics
Social media, news, blog, policy document mentions
Citation Metrics
2044
OpenAlex
583
Influential
Cite This
Zachary Teed,
Jia Deng
(2020).
RAFT: Recurrent All-Pairs Field Transforms for Optical Flow.
Lecture notes in computer science
, 402-419.
https://doi.org/10.1007/978-3-030-58536-5_24
Identifiers
- DOI
- 10.1007/978-3-030-58536-5_24
- PMID
- 40686667
- PMCID
- PMC12274184
- arXiv
- 2003.12039