Transform-invariant Convolutional Neural Networks For Image Classification And Search

Abstract

Convolutional neural networks (CNNs) have achieved state-of-the-art results on many visual recognition tasks. However, current CNN models still exhibit a poor ability to be invariant to spatial transformations of images. Intuitively, with sufficient layers and parameters, hierarchical combinations of convolution (matrix multiplication and non-linear activation) and pooling operations should be able to learn a robust mapping from transformed input images to transform-invariant representations. In this paper, we propose randomly transforming (rotation, scale, and translation) feature maps of CNNs during the training stage. This prevents complex dependencies of specific rotation, scale, and translation levels of training images in CNN models. Rather, each convolutional kernel learns to detect a feature that is generally helpful for producing the transform-invariant answer given the combinatorially large variety of transform levels of its input feature maps. In this way, we do not require

Stats

• citations31
• S2 citations—
• github stars13
• HF likes0
• heat score13.58
• arxiv keyshen2019transform

Related papers

• Patch Reordering: A Novel Way To Achieve Rotation And Translation Invariance In Convolutional Neural Networks (2019)4.55
• Group Invariant Deep Representations For Image Instance Retrieval (2016)0.00
• Compensating For Large In-plane Rotations In Natural Images (2016)6.34
• Rotation Invariant Deep CBIR (2020)0.00
• Volumetric Transformer Networks (2020)4.52
• Class-weighted Convolutional Features For Visual Instance Search (2017)12.81
• From Selective Deep Convolutional Features To Compact Binary Representations For Image Retrieval (2018)10.35
• Co-occurrence Of Deep Convolutional Features For Image Search (2020)9.76