Oriented Response Networks

By Yanzhao Zhou, Qixiang Ye, Qiang Qiu, Jianbin Jiao. 2017

283 citations recorded by the source corpus.

Deep Convolution Neural Networks (DCNNs) are capable of learning unprecedentedly effective image representations. However, their ability in handling significant local and global image rotations remains limited. In this paper, we propose Active Rotating Filters (ARFs) that actively rotate during convolution and produce feature maps with location and orientation explicitly encoded. An ARF acts as a virtual filter bank containing the filter itself and its multiple unmaterialised rotated versions. During back-propagation, an ARF is collectively updated using errors from all its rotated versions. DCNNs using ARFs, referred to as Oriented Response Networks (ORNs), can produce within-class rotation-invariant deep features while maintaining inter-class discrimination for classification tasks. The oriented response produced by ORNs can also be used for image and object orientation estimation tasks. Over multiple state-of-the-art DCNN architectures, such as VGG, ResNet, and STN, we consistently observe that replacing regular filters with the proposed ARFs leads to significant reduction in the number of network parameters and improvement in classification performance. We report the best results on several commonly used benchmarks.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

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

In Aerospace Engineering

In Computer Vision and Pattern Recognition

In Model Architectures

Citations in the corpus

Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data (2020, methodology): “This approach is easy to implement, and has also been used when the feature maps are vector ﬁelds (Zhou et al., 2017; Marcos et al., 2017), and with other representations (Cohen and Welling, 2016b), but only on image data where locations are discrete and the group cardinality is small.”
Tensor Field Networks: Rotation- and Translation-Equivariant Neural Networks for 3D Point Clouds (2018, background): “Other authors have investigated the problems of rotation equivariance in 2D, such as Zhou et al. [4], Gonzalez et al. [5], Li et al. [6].”
Learning SO(3) Equivariant Representations with Spherical CNNs (2017, neutral): “[17] and Marcos et al .”
Rotation Equivariant Vector Field Networks (2016, methodology): “01%, thus improving significantly over TIPooling and over the more recent H-Net [29] and ORN [31].”
Polar Transformer Networks (2017, background): “[30] applies rotating filters that produce explicitly oriented feature maps to achieve rotation invariant features.”
CubeNet: Equivariance to 3D Rotation and Translation (2018, methodology): “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].”
ReDet: A Rotation-equivariant Detector for Aerial Object Detection (2021, methodology): “The orientation pooling operation is usually adopted in classification tasks [5, 33, 45].”
Group Equivariant Capsule Networks (2018, background): “One way is to share weights over differently rotated ﬁlters or augment the input heavily by transformations [Yanzhao et al., 2017, Weiler et al., 2018].”
B-Spline CNNs on Lie Groups (2019, background): “…& Brostow, 2018; Hoogeboom et al., 2018), regular continuous G-CNNs (Oyallon & Mallat, 2015; Bekkers et al., 2015a; 2018b; Weiler et al., 2018a; Zhou et al., 2017; Marcos et al., 2017) and steerable continuous G-CNNs (Cohen et al., 2018b; Worrall et al., 2017; Kondor & Trivedi, 2018; Thomas et…”
HexaConv (2018, background): “…or the feature maps followed and subsequent channel permutations to obtain equivariant (or invariant) CNNs (Cohen & Welling, 2016; Dieleman et al., 2016; Zhou et al., 2017; Li et al., 2017 To simplify the exposition, we assume there is only one input channel and one output channel; K = C = 1 .”
Dynamic Refinement Network for Oriented and Densely Packed Object Detection (2020, background): “ORN [45] proposes active rotating filters which actively rotate during convolution.”
Sparsity Invariant CNNs (2017, background): “More recently, geometric invariances (e.g., rotation, perspective transformation) have been incorporated directly into the ﬁlters of CNNs [4, 25, 29, 66, 69].”
Learning a Rotation Invariant Detector with Rotatable Bounding Box (2017, result): “Zhou et al . proposed a network named ORN [27] which shows similar character with our method.”
Land cover mapping at very high resolution with rotation equivariant CNNs: towards small yet accurate models (2018, methodology): “Cohen and Welling (2016) encode equivariance to shifts and to rotations by multiples of 90o by tying filter weights, while Zhou et al. (2017) use linearly interpolated filters.”
RepPoints: Point Set Representation for Object Detection (2019, background): “These include an elliptic representation for pedestrian detection [22] and a rotated bounding box to better handle rotational variations [16, 48].”
Gabor Convolutional Networks (2017, methodology): “In contrast, ORNs only actively rotate the ﬁlters and lack a feature enhancement process.”
Deep Learning for Generic Object Detection: A Survey (2018, background): “refore, many approaches have been presented to enhance robustness, aiming at learning invariant CNN representations with respect to different types of transformations such as scale [131,21], rotation [21,42,284,323], or both [126]. 20 Li Liu et al. One representative work is Spatial Transformer Netw”