Scale Equivariant CNNs with Scale Steerable Filters

Convolution Neural Networks (CNNs), despite being one of the most successful image classification methods, are not robust to most geometric transformations (rotation, isotropic scaling) because of their structural constraints. Recently, scale steerable filters have been proposed to allow scale invariance in CNNs. Although these filters enhance the network performance in scaled image classification tasks, they cannot maintain the scale information across the network. In this paper, this problem is addressed. First, a CNN is built with the usage of scale steerable filters. Then, a scale equivariat network is acquired by adding a feature map to each layer so that the scale-related features are retained across the network. At last, by defining the cost function as the cross entropy, this solution is evaluated and the model parameters are updated. The results show that it improves the perfromance about 2% over other comparable methods of scale equivariance and scale invariance, when run on the FMNIST-scale dataset.

[1]  Max Welling,et al.  Group Equivariant Convolutional Networks , 2016, ICML.

[2]  Zhitao Gong,et al.  Strike (With) a Pose: Neural Networks Are Easily Fooled by Strange Poses of Familiar Objects , 2018, 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[3]  Roland Vollgraf,et al.  Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms , 2017, ArXiv.

[4]  Nikos Komodakis,et al.  Rotation Equivariant Vector Field Networks , 2016, 2017 IEEE International Conference on Computer Vision (ICCV).

[5]  Jiaxing Zhang,et al.  Scale-Invariant Convolutional Neural Networks , 2014, ArXiv.

[6]  Maurice Weiler,et al.  A General Theory of Equivariant CNNs on Homogeneous Spaces , 2018, NeurIPS.

[7]  Daniel E. Worrall,et al.  Deep Scale-spaces: Equivariance Over Scale , 2019, NeurIPS.

[8]  A. Robert Calderbank,et al.  Scale-Equivariant Neural Networks with Decomposed Convolutional Filters , 2019, ArXiv.

[9]  Maurice Weiler,et al.  Learning Steerable Filters for Rotation Equivariant CNNs , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[10]  Jonathon Shlens,et al.  Explaining and Harnessing Adversarial Examples , 2014, ICLR.

[11]  Takashi Matsubara,et al.  Scale-Invariant Recognition by Weight-Shared CNNs in Parallel , 2017, ACML.

[12]  Devis Tuia,et al.  Scale equivariance in CNNs with vector fields , 2018, ArXiv.

[13]  Edward H. Adelson,et al.  The Design and Use of Steerable Filters , 1991, IEEE Trans. Pattern Anal. Mach. Intell..

[14]  Mingyan Liu,et al.  Spatially Transformed Adversarial Examples , 2018, ICLR.

[15]  Andrea Vedaldi,et al.  Understanding Image Representations by Measuring Their Equivariance and Equivalence , 2014, International Journal of Computer Vision.

[16]  Andrea Vedaldi,et al.  Warped Convolutions: Efficient Invariance to Spatial Transformations , 2016, ICML.

[17]  Max Welling,et al.  Steerable CNNs , 2016, ICLR.

[18]  Anupam K. Gupta,et al.  Scale Steerable Filters for Locally Scale-Invariant Convolutional Neural Networks , 2019, ArXiv.

[19]  Ivan Sosnovik,et al.  Scale-Equivariant Steerable Networks , 2020, ICLR.

[20]  David W. Jacobs,et al.  Locally Scale-Invariant Convolutional Neural Networks , 2014, ArXiv.

[21]  Joan Bruna,et al.  Intriguing properties of neural networks , 2013, ICLR.

[22]  Stephan J. Garbin,et al.  Harmonic Networks: Deep Translation and Rotation Equivariance , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[23]  Atul Prakash,et al.  Robust Physical-World Attacks on Deep Learning Visual Classification , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[24]  Stéphane Mallat,et al.  Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination , 2013, 2013 IEEE Conference on Computer Vision and Pattern Recognition.