Dynamic Steerable Blocks in Deep Residual Networks

By Jörn-Henrik Jacobsen, Bert De Brabandere, Arnold M. W. Smeulders. 2017

28 citations recorded by the source corpus.

Filters in convolutional networks are typically parameterized in a pixel basis, that does not take prior knowledge about the visual world into account. We investigate the generalized notion of frames designed with image properties in mind, as alternatives to this parametrization. We show that frame-based ResNets and Densenets can improve performance on Cifar-10+ consistently, while having additional pleasant properties like steerability. By exploiting these transformation properties explicitly, we arrive at dynamic steerable blocks. They are an extension of residual blocks, that are able to seamlessly transform filters under pre-defined transformations, conditioned on the input at training and inference time. Dynamic steerable blocks learn the degree of invariance from data and locally adapt filters, allowing them to apply a different geometrical variant of the same filter to each location of the feature map. When evaluated on the Berkeley Segmentation contour detection dataset, our approach outperforms all competing approaches that do not utilize pre-training. Our results highlight the benefits of image-based regularization to deep networks.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Stacked Capsule Autoencoders (2019, background): “Spatial Transformers ( ST ; Jaderberg et al., 2015) apply afﬁne transformations to the image sampling grid while steerable networks (Cohen and Welling, 2017; Jacobsen et al., 2017) dynamically change convolutional ﬁlters.”
Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data (2020, background): “Another approach, drawing on harmonic analysis, finds a basis of equivariant functions and parametrizes convolutional kernels in that basis (Worrall et al., 2017; Weiler and Cesa, 2019; Jacobsen et al., 2017).”
Learning Steerable Filters for Rotation Equivariant CNNs (2017, background): “In [18] transformations are dealt with by convolving with filters which are steered by a subnetwork.”
CubeNet: Equivariance to 3D Rotation and Translation (2018, background): “Recently there has been an explosion of interest into CNNs with predefined transformation equivariances, beyond translation [8, 50, 11, 29, 16, 42, 36, 14, 19, 31, 18, 22, 15, 55, 49, 26, 25, 48, 33, 28, 9].”
Convolutional Networks for Spherical Signals (2017, methodology): “Successful implementations of equivariant neural networks include (Worrall et al., 2016; Jacobsen et al., 2017; Zhou et al., 2017; Ravanbakhsh et al., 2016; Dieleman et al., 2016; Cohen & Welling, 2016; 2017).”