Paper Info
Title
Kernel Transformer Networks For Compact Spherical Convolution
Abstract
Ideally, 360 degrees imagery could inherit the deep convolutional neural networks (CNNs) already trained with great success on perspective projection images. However, existing methods to transfer CNNs from perspective to spherical images introduce significant computational costs and/or degradations in accuracy. We present the Kernel Transformer Network (KTN) to efficiently transfer convolution kernels from perspective images to the equirectangular projection of 360 degrees images. Given a source CNN for perspective images as input, the KTN produces a function parameterized by a polar angle and kernel as output. Given a novel 360 degrees image, that function in turn can compute convolutions for arbitrary layers and kernels as would the source CNN on the corresponding tangent plane projections. Distinct from all existing methods, KTNs allow model transfer: the same model can be applied to different source CNNs with the same base architecture. This enables application to multiple recognition tasks without re-training the KTN. Validating our approach with multiple source CNNs and datasets, we show that KTNs improve the state of the art for spherical convolution. KTNs successfully preserve the source CNN's accuracy, while offering transferability, scalability to typical image resolutions, and, in many cases, a substantially lower memory footprint(1).
Year
DOI
Venue
2018
10.1109/CVPR.2019.00967
2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2019)
Field
DocType
Volume
Kernel (linear algebra),Parameterized complexity,Pattern recognition,Convolutional neural network,Convolution,Computer science,Equirectangular projection,Algorithm,Polar coordinate system,Perspective (graphical),Artificial intelligence,Image resolution
Journal
abs/1812.03115
ISSN
Citations 
PageRank 
1063-6919
3
0.40
References 
Authors
0
2
Name
Order
Citations
PageRank
Yu-Chuan Su18714.90
Kristen Grauman26258326.34