Proceedings Article10.1109/IRC.2019.00090
Deep Rotating Kernel Convolution Neural Network
Crino Shin,Jongpil Yun +1 more
- 01 Feb 2019
- pp 441-442
3
TL;DR: A simple and highly scalable model that has excellent rotational invariant characteristics by using Rotating Kernel Convolution (RK Conv) which convolves and rotates kernel and Global Average Pooling which invariant features to absolute position is proposed.
read more
Abstract: This paper describes a method that can be efficiently applied to data with rotational invariant characteristics such as texture. We propose a simple and highly scalable model that has excellent rotational invariant characteristics by using Rotating Kernel Convolution(RK Conv) which convolves and rotates kernel and Global Average Pooling (GAP) which invariant features to absolute position. The proposed model shows the state of the art performance in experiments under the same conditions as those in previous papers.
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Citations
Rotation Invariant Networks for Image Classification for HPC and Embedded Systems
TL;DR: This work presents the next step to obtain a general solution to endowing CNN architectures with the capability of classifying rotated objects and predicting the rotation angle without data-augmentation techniques and presents results obtained using a Gabor-filter bank and a ResNet feature backbone.
5
Rotation-Invariant Descriptors Learned with Circulant Convolution Neural Networks
Wenwei Lin,Chonghao Zhong,Xunpei Sun,Haitao Meng,Gang Chen,Biao Hu,Zonghua Gu +6 more
- 06 Nov 2023
TL;DR: RICNN, a novel deep learning framework that encodes invariance against the rotations in geometry explicitly into convolutional neural networks, and proposes a novel multi-level hinge triplet loss function to strengthen the matching constraints against geometry rotations.
A minimal model for classification of rotated objects with prediction of the angle of rotation
TL;DR: In this article, a new reduced rotation invariant classification model composed of two parts: a feature representation mapping and a classifier is proposed, which can predict the rotation angle and achieve state-of-the-art performance on MNIST and CIFAR-10 datasets.
References
•Proceedings Article
Spatial transformer networks
Max Jaderberg,Karen Simonyan,Andrew Zisserman,Koray Kavukcuoglu +3 more
- 07 Dec 2015
TL;DR: This work introduces a new learnable module, the Spatial Transformer, which explicitly allows the spatial manipulation of data within the network, and can be inserted into existing convolutional architectures, giving neural networks the ability to actively spatially transform feature maps.
Oriented Response Networks
Yanzhao Zhou,Qixiang Ye,Qiang Qiu,Jianbin Jiao +3 more
- 01 Jul 2017
TL;DR: Active rotating filters (ARFs) as mentioned in this paper can be used to produce within-class rotation-invariant deep features while maintaining inter-class discrimination for classification tasks, which can also be used for image and object orientation estimation.
Rotation Invariant Local Binary Convolution Neural Networks
Xin Zhang,Li Liu,Yuxiang Xie,Jie Chen,Lingda Wu,Matti Pietikäinen +5 more
- 01 Oct 2017
TL;DR: A new convolutional module, local binary orientation module (LBoM), which takes advantages of both local binary convolutionals and active rotating filters to effectively deal with the rotation variations with fewer parameters is proposed.
A Rotationally-Invariant Convolution Module by Feature Map Back-Rotation
Patrick Follmann,Tobias Böttger +1 more
- 12 Mar 2018
TL;DR: This work uses rotational convolutions and introduces a rotational pooling layer that performs a pooling over the back-rotated output feature maps that allows to train on unrotated data and perform well on a rotated test set.
38
Rotation Invariant Local Binary Convolution Neural Networks
TL;DR: Through replacing the basic convolution layer in DCNN with LBoMs, RI-LBCNN can be easily implemented and LBoM can be naturally inserted to other popular models without any extra modification to the optimisation process.