Learning Energy Based Inpainting for Optical Flow

Research output: Contribution to conferencePaper

Abstract

Modern optical flow methods are often composed of a cascade of many independent steps or formulated as a black box neural network that is hard to interpret and analyze. In this work we seek for a plain, interpretable, but learnable solution. We propose a novel inpainting based algorithm that approaches the problem in three steps: feature selection and matching, selection of supporting points and energy based inpainting. To facilitate the inference we propose an optimization layer that allows to backpropagate through 10K iterations of a first-order method without any numerical or memory problems. Compared to recent state-of-the-art networks, our modular CNN is very lightweight and competitive with other, more involved, inpainting based methods.

Conference

Conference14th Asian Conference on Computer Vision
Abbreviated titleACCV
CountryAustralia
CityPerth
Period4/12/186/12/18
Internet address

Fingerprint

Optical flows
Feature extraction
Neural networks
Data storage equipment

Keywords

  • Optical Flow
  • Energy Optimization
  • Deep Learning

Cite this

Vogel, C., Knöbelreiter, P., & Pock, T. (2018). Learning Energy Based Inpainting for Optical Flow. Paper presented at 14th Asian Conference on Computer Vision, Perth, Australia.

Learning Energy Based Inpainting for Optical Flow. / Vogel, Christoph; Knöbelreiter, Patrick; Pock, Thomas.

2018. Paper presented at 14th Asian Conference on Computer Vision, Perth, Australia.

Research output: Contribution to conferencePaper

Vogel, C, Knöbelreiter, P & Pock, T 2018, 'Learning Energy Based Inpainting for Optical Flow' Paper presented at 14th Asian Conference on Computer Vision, Perth, Australia, 4/12/18 - 6/12/18, .
Vogel C, Knöbelreiter P, Pock T. Learning Energy Based Inpainting for Optical Flow. 2018. Paper presented at 14th Asian Conference on Computer Vision, Perth, Australia.
Vogel, Christoph ; Knöbelreiter, Patrick ; Pock, Thomas. / Learning Energy Based Inpainting for Optical Flow. Paper presented at 14th Asian Conference on Computer Vision, Perth, Australia.
@conference{88811ea9f7f84656a7a2d93ee60ff201,
title = "Learning Energy Based Inpainting for Optical Flow",
abstract = "Modern optical flow methods are often composed of a cascade of many independent steps or formulated as a black box neural network that is hard to interpret and analyze. In this work we seek for a plain, interpretable, but learnable solution. We propose a novel inpainting based algorithm that approaches the problem in three steps: feature selection and matching, selection of supporting points and energy based inpainting. To facilitate the inference we propose an optimization layer that allows to backpropagate through 10K iterations of a first-order method without any numerical or memory problems. Compared to recent state-of-the-art networks, our modular CNN is very lightweight and competitive with other, more involved, inpainting based methods.",
keywords = "Optical Flow, Energy Optimization, Deep Learning",
author = "Christoph Vogel and Patrick Kn{\"o}belreiter and Thomas Pock",
year = "2018",
month = "12",
day = "4",
language = "English",
note = "14th Asian Conference on Computer Vision, ACCV ; Conference date: 04-12-2018 Through 06-12-2018",
url = "http://accv2018.net",

}

TY - CONF

T1 - Learning Energy Based Inpainting for Optical Flow

AU - Vogel,Christoph

AU - Knöbelreiter,Patrick

AU - Pock,Thomas

PY - 2018/12/4

Y1 - 2018/12/4

N2 - Modern optical flow methods are often composed of a cascade of many independent steps or formulated as a black box neural network that is hard to interpret and analyze. In this work we seek for a plain, interpretable, but learnable solution. We propose a novel inpainting based algorithm that approaches the problem in three steps: feature selection and matching, selection of supporting points and energy based inpainting. To facilitate the inference we propose an optimization layer that allows to backpropagate through 10K iterations of a first-order method without any numerical or memory problems. Compared to recent state-of-the-art networks, our modular CNN is very lightweight and competitive with other, more involved, inpainting based methods.

AB - Modern optical flow methods are often composed of a cascade of many independent steps or formulated as a black box neural network that is hard to interpret and analyze. In this work we seek for a plain, interpretable, but learnable solution. We propose a novel inpainting based algorithm that approaches the problem in three steps: feature selection and matching, selection of supporting points and energy based inpainting. To facilitate the inference we propose an optimization layer that allows to backpropagate through 10K iterations of a first-order method without any numerical or memory problems. Compared to recent state-of-the-art networks, our modular CNN is very lightweight and competitive with other, more involved, inpainting based methods.

KW - Optical Flow

KW - Energy Optimization

KW - Deep Learning

M3 - Paper

ER -