Please use this identifier to cite or link to this item:
http://hdl.handle.net/1942/40111
Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Fang, Guibiao | - |
dc.contributor.author | CHEN, Junhong | - |
dc.contributor.author | Liang, Dayong | - |
dc.contributor.author | Asim, Muhammad | - |
dc.contributor.author | VAN REETH, Frank | - |
dc.contributor.author | CLAESEN, Luc | - |
dc.contributor.author | Yang, Zhenguo | - |
dc.contributor.author | Liu, Wenyin | - |
dc.date.accessioned | 2023-05-15T12:05:29Z | - |
dc.date.available | 2023-05-15T12:05:29Z | - |
dc.date.issued | 2023 | - |
dc.date.submitted | 2023-05-07T14:32:24Z | - |
dc.identifier.citation | NEURAL PROCESSING LETTERS, | - |
dc.identifier.issn | 1370-4621 | - |
dc.identifier.uri | http://hdl.handle.net/1942/40111 | - |
dc.description.abstract | Cost volume is widely used to establish correspondences in optical flow estimation. However, when dealing with low-texture and occluded areas, it is difficult to estimate the cost volume correctly. Therefore, we propose a replacement: feature correlation transformer (FCTR), a transformer with self-and cross-attention alternations for obtaining global receptive fields and positional embedding for establishing correspondences. With global context and posi-tional information, FCTR can produce more accurate correspondences for ambiguous areas. Using position-embedded feature allows the removal of the context network; the positional information can be aggregated within ambiguous motion boundaries, and the number of model parameters can be reduced. To speed up network convergence and strengthen robust-ness, we introduce a smooth L1 loss with exponential weights in the pre-training step. At the time of submission, our method achieves competitive performance with all published optical flow methods on both the KITTI-2015 and MPI-Sintel benchmarks. Moreover, it outperforms all optical flow and scene flow methods in KITTI-2015 foreground-region prediction. | - |
dc.language.iso | en | - |
dc.publisher | Springer Science and Business Media {LLC} | - |
dc.subject.other | Optical flow | - |
dc.subject.other | Cost volume | - |
dc.subject.other | Ambiguous correspondence | - |
dc.subject.other | Transformer | - |
dc.subject.other | Alternating attention | - |
dc.title | Feature Correlation Transformer for Estimating Ambiguous Optical Flow | - |
dc.type | Journal Contribution | - |
local.bibliographicCitation.jcat | A1 | - |
local.publisher.place | VAN GODEWIJCKSTRAAT 30, 3311 GZ DORDRECHT, NETHERLANDS | - |
local.type.refereed | Refereed | - |
local.type.specified | Article | - |
local.bibliographicCitation.status | Early view | - |
dc.identifier.doi | 10.1007/s11063-023-11273-6 | - |
dc.identifier.isi | 000982933300003 | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.contributor.orcid | null | - |
dc.identifier.eissn | 1573-773X | - |
local.provider.type | Orcid | - |
local.uhasselt.international | yes | - |
item.fulltext | With Fulltext | - |
item.accessRights | Open Access | - |
item.fullcitation | Fang, Guibiao; CHEN, Junhong; Liang, Dayong; Asim, Muhammad; VAN REETH, Frank; CLAESEN, Luc; Yang, Zhenguo & Liu, Wenyin (2023) Feature Correlation Transformer for Estimating Ambiguous Optical Flow. In: NEURAL PROCESSING LETTERS,. | - |
item.contributor | Fang, Guibiao | - |
item.contributor | CHEN, Junhong | - |
item.contributor | Liang, Dayong | - |
item.contributor | Asim, Muhammad | - |
item.contributor | VAN REETH, Frank | - |
item.contributor | CLAESEN, Luc | - |
item.contributor | Yang, Zhenguo | - |
item.contributor | Liu, Wenyin | - |
crisitem.journal.issn | 1370-4621 | - |
crisitem.journal.eissn | 1573-773X | - |
Appears in Collections: | Research publications |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
s11063-023-11273-6.pdf Restricted Access | Early view | 2.56 MB | Adobe PDF | View/Open Request a copy |
FCTR(ral)-Final-提交版.pdf | Peer-reviewed author version | 3.79 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.