Convolutional neural networks (CNNs) have recently been very successful in a
variety of computer vision tasks, especially on those linked to recognition.
Optical flow estimation has not been among the tasks where CNNs were
successful. In this paper we construct appropriate CNNs which are capable of
solving the optical flow estimation problem as a supervised learning task. We
propose and compare two architectures: a generic architecture and another one
including a layer that correlates feature vectors at different image locations.
Since existing ground truth data sets are not sufficiently large to train a
CNN, we generate a synthetic Flying Chairs dataset. We show that networks
trained on this unrealistic data still generalize very well to existing
datasets such as Sintel and KITTI, achieving competitive accuracy at frame
rates of 5 to 10 fps.
Description
FlowNet: Learning Optical Flow with Convolutional Networks
%0 Generic
%1 fischer2015flownet
%A Fischer, Philipp
%A Dosovitskiy, Alexey
%A Ilg, Eddy
%A Häusser, Philip
%A Hazırbaş, Caner
%A Golkov, Vladimir
%A van der Smagt, Patrick
%A Cremers, Daniel
%A Brox, Thomas
%D 2015
%K deep-learning
%T FlowNet: Learning Optical Flow with Convolutional Networks
%U http://arxiv.org/abs/1504.06852
%X Convolutional neural networks (CNNs) have recently been very successful in a
variety of computer vision tasks, especially on those linked to recognition.
Optical flow estimation has not been among the tasks where CNNs were
successful. In this paper we construct appropriate CNNs which are capable of
solving the optical flow estimation problem as a supervised learning task. We
propose and compare two architectures: a generic architecture and another one
including a layer that correlates feature vectors at different image locations.
Since existing ground truth data sets are not sufficiently large to train a
CNN, we generate a synthetic Flying Chairs dataset. We show that networks
trained on this unrealistic data still generalize very well to existing
datasets such as Sintel and KITTI, achieving competitive accuracy at frame
rates of 5 to 10 fps.
@misc{fischer2015flownet,
abstract = {Convolutional neural networks (CNNs) have recently been very successful in a
variety of computer vision tasks, especially on those linked to recognition.
Optical flow estimation has not been among the tasks where CNNs were
successful. In this paper we construct appropriate CNNs which are capable of
solving the optical flow estimation problem as a supervised learning task. We
propose and compare two architectures: a generic architecture and another one
including a layer that correlates feature vectors at different image locations.
Since existing ground truth data sets are not sufficiently large to train a
CNN, we generate a synthetic Flying Chairs dataset. We show that networks
trained on this unrealistic data still generalize very well to existing
datasets such as Sintel and KITTI, achieving competitive accuracy at frame
rates of 5 to 10 fps.},
added-at = {2017-08-17T09:35:04.000+0200},
author = {Fischer, Philipp and Dosovitskiy, Alexey and Ilg, Eddy and Häusser, Philip and Hazırbaş, Caner and Golkov, Vladimir and van der Smagt, Patrick and Cremers, Daniel and Brox, Thomas},
biburl = {https://www.bibsonomy.org/bibtex/2f9d969dfcf5f3e3d574206a2ffc1e756/axel.vogler},
description = {FlowNet: Learning Optical Flow with Convolutional Networks},
interhash = {cada35d856aea633c31f5c9a80e7acf2},
intrahash = {f9d969dfcf5f3e3d574206a2ffc1e756},
keywords = {deep-learning},
note = {cite arxiv:1504.06852Comment: Added supplementary material},
timestamp = {2017-08-17T09:35:04.000+0200},
title = {FlowNet: Learning Optical Flow with Convolutional Networks},
url = {http://arxiv.org/abs/1504.06852},
year = 2015
}