Locally Scale-Invariant Convolutional Neural Networks

Abstract

Convolutional Neural Networks (ConvNets) have shown excellent results on many visual classification tasks. With the exception of ImageNet, these datasets are carefully crafted such that objects are well-aligned at similar scales. Naturally, the feature learning problem gets more challenging as the amount of variation in the data increases, as the models have to learn to be invariant to certain changes in appearance. Recent results on the ImageNet dataset show that given enough data, ConvNets can learn such invariances producing very discriminative features [1]. But could we do more: use less parameters, less data, learn more discriminative features, if certain invariances were built into the learning process? In this paper we present a simple model that allows ConvNets to learn features in a locally scale-invariant manner without increasing the number of model parameters. We show on a modified MNIST dataset that when faced with scale variation, building in scale-invariance allows ConvNets to learn more discriminative features with reduced chances of over-fitting.

Deep Learning and Representation Learning Workshop: NIPS 2014

References in corpus (1)

• Learning Invariant Representations with Local Transformations

Cited by in corpus (44)

• Universal Correspondence Network
• Deep Learning for Generic Object Detection: A Survey
• Spatial Transformer Networks
• Deep High-Resolution Representation Learning for Human Pose Estimation
• ZNN - A Fast and Scalable Algorithm for Training 3D Convolutional Networks on Multi-Core and Many-Core Shared Memory Machines
• Scale-Equivariant Steerable Networks
• Scale equivariance in CNNs with vector fields
• Interaction-and-Aggregation Network for Person Re-identification
• Scale Steerable Filters for Locally Scale-Invariant Convolutional Neural Networks
• Deformable Kernels: Adapting Effective Receptive Fields for Object Deformation
• Learning Deep NBNN Representations for Robust Place Categorization
• Neural Networks Enforcing Physical Symmetries in Nonlinear Dynamical Lattices: The Case Example of the Ablowitz-Ladik Model
• Resolution learning in deep convolutional networks using scale-space theory
• Warped Convolutions: Efficient Invariance to Spatial Transformations
• Self-supervised Scale Equivariant Network for Weakly Supervised Semantic Segmentation
• Blurring the Line Between Structure and Learning to Optimize and Adapt Receptive Fields
• Scale-invariant scale-channel networks: Deep networks that generalise to previously unseen scales
• Dreaming More Data: Class-dependent Distributions over Diffeomorphisms for Learned Data Augmentation
• Scaling-Translation-Equivariant Networks with Decomposed Convolutional Filters
• Learning Steerable Filters for Rotation Equivariant CNNs
• Understanding Important Features of Deep Learning Models for Transmission Electron Microscopy Image Segmentation
• Towards Learning Affine-Invariant Representations via Data-Efficient CNNs
• DeSTNet: Densely Fused Spatial Transformer Networks
• Dynamic Scale Inference by Entropy Minimization
• Temporal Transformer Networks: Joint Learning of Invariant and Discriminative Time Warping
• Hierarchical Spatial Transformer Network
• Downbeat Tracking with Tempo-Invariant Convolutional Neural Networks
• Sampling Equivariant Self-attention Networks for Object Detection in Aerial Images
• Learnable Gabor modulated complex-valued networks for orientation robustness
• Learning to Point and Count
• Master's Thesis : Deep Learning for Visual Recognition
• Deformation Robust Roto-Scale-Translation Equivariant CNNs
• Euclidean Invariant Recognition of 2D Shapes Using Histograms of Magnitudes of Local Fourier-Mellin Descriptors
• Do CNNs Encode Data Augmentations?
• Discriminative Learning of Similarity and Group Equivariant Representations
• Learning scale-variant and scale-invariant features for deep image classification
• Affine Disentangled GAN for Interpretable and Robust AV Perception
• A Progressively-trained Scale-invariant and Boundary-aware Deep Neural Network for the Automatic 3D Segmentation of Lung Lesions
• A Closed-Form Learned Pooling for Deep Classification Networks
• Wiggling Weights to Improve the Robustness of Classifiers
• Group Equivariant Subsampling
• Quantised Transforming Auto-Encoders: Achieving Equivariance to Arbitrary Transformations in Deep Networks
• Patch Reordering: a Novel Way to Achieve Rotation and Translation Invariance in Convolutional Neural Networks
• Transform-Invariant Convolutional Neural Networks for Image Classification and Search