Convolutional neural networks (CNNs) are inherently limited to model
geometric transformations due to the fixed geometric structures in its building
modules. In this work, we introduce two new modules to enhance the
transformation modeling capacity of CNNs, namely, deformable convolution and
deformable RoI pooling. Both are based on the idea of augmenting the spatial
sampling locations in the modules with additional offsets and learning the
offsets from target tasks, without additional supervision. The new modules can
readily replace their plain counterparts in existing CNNs and can be easily
trained end-to-end by standard back-propagation, giving rise to deformable
convolutional networks. Extensive experiments validate the effectiveness of our
approach on sophisticated vision tasks of object detection and semantic
segmentation. The code would be released.
Описание
Deformable convolution is a convolution where the kernel-elements have an offset
%0 Generic
%1 dai2017deformable
%A Dai, Jifeng
%A Qi, Haozhi
%A Xiong, Yuwen
%A Li, Yi
%A Zhang, Guodong
%A Hu, Han
%A Wei, Yichen
%D 2017
%K conv-nets deformable-convolution
%T Deformable Convolutional Networks
%U http://arxiv.org/abs/1703.06211
%X Convolutional neural networks (CNNs) are inherently limited to model
geometric transformations due to the fixed geometric structures in its building
modules. In this work, we introduce two new modules to enhance the
transformation modeling capacity of CNNs, namely, deformable convolution and
deformable RoI pooling. Both are based on the idea of augmenting the spatial
sampling locations in the modules with additional offsets and learning the
offsets from target tasks, without additional supervision. The new modules can
readily replace their plain counterparts in existing CNNs and can be easily
trained end-to-end by standard back-propagation, giving rise to deformable
convolutional networks. Extensive experiments validate the effectiveness of our
approach on sophisticated vision tasks of object detection and semantic
segmentation. The code would be released.
@misc{dai2017deformable,
abstract = {Convolutional neural networks (CNNs) are inherently limited to model
geometric transformations due to the fixed geometric structures in its building
modules. In this work, we introduce two new modules to enhance the
transformation modeling capacity of CNNs, namely, deformable convolution and
deformable RoI pooling. Both are based on the idea of augmenting the spatial
sampling locations in the modules with additional offsets and learning the
offsets from target tasks, without additional supervision. The new modules can
readily replace their plain counterparts in existing CNNs and can be easily
trained end-to-end by standard back-propagation, giving rise to deformable
convolutional networks. Extensive experiments validate the effectiveness of our
approach on sophisticated vision tasks of object detection and semantic
segmentation. The code would be released.},
added-at = {2022-12-14T17:34:20.000+0100},
author = {Dai, Jifeng and Qi, Haozhi and Xiong, Yuwen and Li, Yi and Zhang, Guodong and Hu, Han and Wei, Yichen},
biburl = {https://www.bibsonomy.org/bibtex/2de3be367bffeed140127fec0f931aca5/geistgesicht},
description = {Deformable convolution is a convolution where the kernel-elements have an offset},
interhash = {cb47166efb02b866f28e7892d5c8a02b},
intrahash = {de3be367bffeed140127fec0f931aca5},
keywords = {conv-nets deformable-convolution},
note = {cite arxiv:1703.06211},
timestamp = {2022-12-14T17:34:20.000+0100},
title = {Deformable Convolutional Networks},
url = {http://arxiv.org/abs/1703.06211},
year = 2017
}