IEICE TRANSACTIONS on Fundamentals
OFR-Net: Optical Flow Refinement with a Pyramid Dense Residual Network
Summary :
This paper proposes a new and effective convolutional neural network model termed OFR-Net for optical flow refinement. The OFR-Net exploits the spatial correlation between images and optical flow fields. It adopts a pyramidal codec structure with residual connections, dense connections and skip connections within and between the encoder and decoder, to comprehensively fuse features of different scales, locally and globally. We also introduce a warp loss to restrict large displacement refinement errors. A series of experiments on the FlyingChairs and MPI Sintel datasets show that the OFR-Net can effectively refine the optical flow predicted by various methods.
- Publication
- IEICE TRANSACTIONS on Fundamentals Vol.E103-A No.11 pp.1312-1318
- Publication Date
- 2020/11/01
- Publicized
- 2020/04/30
- Online ISSN
- 1745-1337
- DOI
- 10.1587/transfun.2020EAL2024
- Type of Manuscript
- LETTER
- Category
- Computer Graphics
Authors
Liping ZHANG
Tsinghua University
Zongqing LU
Tsinghua University
Qingmin LIAO
Tsinghua University
Keyword
Latest Issue
Copyrights notice of machine-translated contents
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Cite this
Copy
Liping ZHANG, Zongqing LU, Qingmin LIAO, "OFR-Net: Optical Flow Refinement with a Pyramid Dense Residual Network" in IEICE TRANSACTIONS on Fundamentals,
vol. E103-A, no. 11, pp. 1312-1318, November 2020, doi: 10.1587/transfun.2020EAL2024.
Abstract: This paper proposes a new and effective convolutional neural network model termed OFR-Net for optical flow refinement. The OFR-Net exploits the spatial correlation between images and optical flow fields. It adopts a pyramidal codec structure with residual connections, dense connections and skip connections within and between the encoder and decoder, to comprehensively fuse features of different scales, locally and globally. We also introduce a warp loss to restrict large displacement refinement errors. A series of experiments on the FlyingChairs and MPI Sintel datasets show that the OFR-Net can effectively refine the optical flow predicted by various methods.
URL: https://global.ieice.org/en_transactions/fundamentals/10.1587/transfun.2020EAL2024/_p
Copy
@ARTICLE{e103-a_11_1312,
author={Liping ZHANG, Zongqing LU, Qingmin LIAO, },
journal={IEICE TRANSACTIONS on Fundamentals},
title={OFR-Net: Optical Flow Refinement with a Pyramid Dense Residual Network},
year={2020},
volume={E103-A},
number={11},
pages={1312-1318},
abstract={This paper proposes a new and effective convolutional neural network model termed OFR-Net for optical flow refinement. The OFR-Net exploits the spatial correlation between images and optical flow fields. It adopts a pyramidal codec structure with residual connections, dense connections and skip connections within and between the encoder and decoder, to comprehensively fuse features of different scales, locally and globally. We also introduce a warp loss to restrict large displacement refinement errors. A series of experiments on the FlyingChairs and MPI Sintel datasets show that the OFR-Net can effectively refine the optical flow predicted by various methods.},
keywords={},
doi={10.1587/transfun.2020EAL2024},
ISSN={1745-1337},
month={November},}
Copy
TY - JOUR
TI - OFR-Net: Optical Flow Refinement with a Pyramid Dense Residual Network
T2 - IEICE TRANSACTIONS on Fundamentals
SP - 1312
EP - 1318
AU - Liping ZHANG
AU - Zongqing LU
AU - Qingmin LIAO
PY - 2020
DO - 10.1587/transfun.2020EAL2024
JO - IEICE TRANSACTIONS on Fundamentals
SN - 1745-1337
VL - E103-A
IS - 11
JA - IEICE TRANSACTIONS on Fundamentals
Y1 - November 2020
AB - This paper proposes a new and effective convolutional neural network model termed OFR-Net for optical flow refinement. The OFR-Net exploits the spatial correlation between images and optical flow fields. It adopts a pyramidal codec structure with residual connections, dense connections and skip connections within and between the encoder and decoder, to comprehensively fuse features of different scales, locally and globally. We also introduce a warp loss to restrict large displacement refinement errors. A series of experiments on the FlyingChairs and MPI Sintel datasets show that the OFR-Net can effectively refine the optical flow predicted by various methods.
ER -
English
