Separable flow: learning motion cost volumes for optical flow estimation
Full-motion cost volumes play a central role in current state-of-the-art optical flow methods. However, constructed using simple feature correlations, they lack the ability to encapsulate prior, or even non-local knowledge. This creates artifacts in poorly constrained ambiguous regions, such as occl...
Main Authors: | , , , |
---|---|
Format: | Conference item |
Language: | English |
Published: |
IEEE
2022
|
_version_ | 1797100263438286848 |
---|---|
author | Zhang, F Woodford, OJ Prisacariu, V Torr, PHS |
author_facet | Zhang, F Woodford, OJ Prisacariu, V Torr, PHS |
author_sort | Zhang, F |
collection | OXFORD |
description | Full-motion cost volumes play a central role in current state-of-the-art optical flow methods. However, constructed using simple feature correlations, they lack the ability to encapsulate prior, or even non-local knowledge. This creates artifacts in poorly constrained ambiguous regions, such as occluded and textureless areas. We propose a separable cost volume module, a drop-in replacement to correlation cost volumes, that uses non-local aggregation layers to exploit global context cues and prior knowledge, in order to disambiguate motions in these regions. Our method leads both the now standard Sintel and KITTI optical flow benchmarks in terms of accuracy, and is also shown to generalize better from synthetic to real data. |
first_indexed | 2024-03-07T05:35:13Z |
format | Conference item |
id | oxford-uuid:e3a886de-f8eb-47d8-b0bf-3ec6d89d4995 |
institution | University of Oxford |
language | English |
last_indexed | 2024-03-07T05:35:13Z |
publishDate | 2022 |
publisher | IEEE |
record_format | dspace |
spelling | oxford-uuid:e3a886de-f8eb-47d8-b0bf-3ec6d89d49952022-03-27T10:10:50ZSeparable flow: learning motion cost volumes for optical flow estimationConference itemhttp://purl.org/coar/resource_type/c_5794uuid:e3a886de-f8eb-47d8-b0bf-3ec6d89d4995EnglishSymplectic ElementsIEEE2022Zhang, FWoodford, OJPrisacariu, VTorr, PHSFull-motion cost volumes play a central role in current state-of-the-art optical flow methods. However, constructed using simple feature correlations, they lack the ability to encapsulate prior, or even non-local knowledge. This creates artifacts in poorly constrained ambiguous regions, such as occluded and textureless areas. We propose a separable cost volume module, a drop-in replacement to correlation cost volumes, that uses non-local aggregation layers to exploit global context cues and prior knowledge, in order to disambiguate motions in these regions. Our method leads both the now standard Sintel and KITTI optical flow benchmarks in terms of accuracy, and is also shown to generalize better from synthetic to real data. |
spellingShingle | Zhang, F Woodford, OJ Prisacariu, V Torr, PHS Separable flow: learning motion cost volumes for optical flow estimation |
title | Separable flow: learning motion cost volumes for optical flow estimation |
title_full | Separable flow: learning motion cost volumes for optical flow estimation |
title_fullStr | Separable flow: learning motion cost volumes for optical flow estimation |
title_full_unstemmed | Separable flow: learning motion cost volumes for optical flow estimation |
title_short | Separable flow: learning motion cost volumes for optical flow estimation |
title_sort | separable flow learning motion cost volumes for optical flow estimation |
work_keys_str_mv | AT zhangf separableflowlearningmotioncostvolumesforopticalflowestimation AT woodfordoj separableflowlearningmotioncostvolumesforopticalflowestimation AT prisacariuv separableflowlearningmotioncostvolumesforopticalflowestimation AT torrphs separableflowlearningmotioncostvolumesforopticalflowestimation |