Local depth edge detection in humans and deep neural networks
Local depth edge detection in humans and deep neural networks
Distinguishing edges caused by a change in depth from other types of edges is an important problem in early vision. We investigate the performance of humans and computer vision models on this task. We use spherical imagery with ground-truth LiDAR range data to build an objective ground-truth dataset for edge classification. We compare various computational models for classifying depth from non-depth edges in small images patches and achieve the best performance (86%) with a convolutional neural network. We investigate human performance on this task in a behavioral experiment and find that human performance is lower than the CNN. Although human and CNN depth responses are correlated, observers’ responses are better predicted by other observers than by the CNN. The responses of CNNs and human observers also show a slightly different pattern of correlation with low-level edge cues, which suggests that CNNs and human observers may weight these features differently for classifying edges.
2681-2689
Ehinger, Krista
3738096b-076a-4137-964e-bdb8a163f9e8
Adams, Wendy
25685aaa-fc54-4d25-8d65-f35f4c5ab688
Graf, Erich
1a5123e2-8f05-4084-a6e6-837dcfc66209
Elder, James H.
f7d4f18e-09dd-4e5c-8fc9-b03064c9ff71
Ehinger, Krista
3738096b-076a-4137-964e-bdb8a163f9e8
Adams, Wendy
25685aaa-fc54-4d25-8d65-f35f4c5ab688
Graf, Erich
1a5123e2-8f05-4084-a6e6-837dcfc66209
Elder, James H.
f7d4f18e-09dd-4e5c-8fc9-b03064c9ff71
Ehinger, Krista, Adams, Wendy, Graf, Erich and Elder, James H.
(2018)
Local depth edge detection in humans and deep neural networks.
In ICCV Workshop on Mutual Benefits of Cognitive and Computer Vision.
IEEE.
.
Record type:
Conference or Workshop Item
(Paper)
Abstract
Distinguishing edges caused by a change in depth from other types of edges is an important problem in early vision. We investigate the performance of humans and computer vision models on this task. We use spherical imagery with ground-truth LiDAR range data to build an objective ground-truth dataset for edge classification. We compare various computational models for classifying depth from non-depth edges in small images patches and achieve the best performance (86%) with a convolutional neural network. We investigate human performance on this task in a behavioral experiment and find that human performance is lower than the CNN. Although human and CNN depth responses are correlated, observers’ responses are better predicted by other observers than by the CNN. The responses of CNNs and human observers also show a slightly different pattern of correlation with low-level edge cues, which suggests that CNNs and human observers may weight these features differently for classifying edges.
Text
EhingerAdamsGrafElder2017
- Accepted Manuscript
More information
Accepted/In Press date: 22 August 2017
e-pub ahead of print date: 23 January 2018
Venue - Dates:
2017 IEEE International on Computer Vision Workshop, , Venice, Italy, 2017-10-22 - 2017-10-29
Identifiers
Local EPrints ID: 413753
URI: http://eprints.soton.ac.uk/id/eprint/413753
ISSN: 2473-9944
PURE UUID: f7379d4f-39ae-4055-8850-a7a9bf1cf587
Catalogue record
Date deposited: 05 Sep 2017 16:30
Last modified: 16 Mar 2024 03:39
Export record
Contributors
Author:
Krista Ehinger
Author:
James H. Elder
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics
