2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017
DOI: 10.1109/cvpr.2017.674
|View full text |Cite
|
Sign up to set email alerts
|

InterpoNet, a Brain Inspired Neural Network for Optical Flow Dense Interpolation

Abstract: Sparse-to-dense interpolation for optical flow is a fundamental phase in the pipeline of most of the leading optical flow estimation algorithms. The current state-of-theart method for interpolation, EpicFlow, is a local average method based on an edge aware geodesic distance. We propose a new data-driven sparse-to-dense interpolation algorithm based on a fully convolutional network. We draw inspiration from the filling-in process in the visual cortex and introduce lateral dependencies between neurons and multi… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
1
1

Citation Types

0
42
0

Year Published

2017
2017
2022
2022

Publication Types

Select...
5
2
1

Relationship

0
8

Authors

Journals

citations
Cited by 43 publications
(42 citation statements)
references
References 42 publications
0
42
0
Order By: Relevance
“…Occlusion has been regarded as an important cue for estimating more accurate optical flow. Because occluded pixels do not have correspondences in the other frame, several approaches [5,12,16,23,32] aim to filter out these outliers to minimize their ill effects and apply post-processing to refine the estimates [22,33,48,67].…”
Section: Related Workmentioning
confidence: 99%
“…Occlusion has been regarded as an important cue for estimating more accurate optical flow. Because occluded pixels do not have correspondences in the other frame, several approaches [5,12,16,23,32] aim to filter out these outliers to minimize their ill effects and apply post-processing to refine the estimates [22,33,48,67].…”
Section: Related Workmentioning
confidence: 99%
“…At first we notice a significant performance drop for both benchmarks, compared to the training set. On KITTI our network outperforms all three inpainting models Epic-Flow [28], [17] and the inpainting network of [41] by a large amount, although all either employ an affine model [28,17] or can learn one [41]. Our method even outperforms [15], who process the cost volume with a MRF model, before using [28] for inpainting.…”
Section: Quantitative Evaluationmentioning
confidence: 91%
“…In contrast to other inpainting based optical flow methods, we start our process from dense matching. Consequently, we are not committed to a pre-selection of, possibly incomplete or unmatchable interest points [37,41], but can select the supporting pixels after matching. Compared to nearest neighbor field methods [18,36], we make use of a complete cost-volume and avoid a coarse-to-fine scheme or hashing.…”
Section: Introduction and Related Workmentioning
confidence: 99%
“…On the contrary, we present an end-to-end approach that performs in-network flow regularization using a flconv layer, which plays a similar role as the regularization term in variational methods. [36]. DeepFlow [24] that involves convolution and pooling operations is however not a CNN, since the "filter weights" are non-trainable image patches.…”
Section: Related Workmentioning
confidence: 99%