Abstract:This work investigates the problem of efficiently learning discriminative low-dimensional (LD) representations of multiclass image objects. We propose a generic end-to-end approach that jointly optimizes sparse dictionary and convolutions for learning LOW-dimensional discriminative image representations, named SparConvLow, taking advantage of convolutional neural networks (CNNs), dictionary learning, and orthogonal projections. The whole learning process can be summarized as follows. First, a CNN module is employed to extract high-dimensional (HD) preliminary convolutional features. Second, to avoid the high computational cost of direct sparse coding on HD CNN features, we learn sparse representation (SR) over a task-driven dictionary in the space with the feature being orthogonally projected. We then exploit the discriminative projection on SR. The whole learning process is consistently treated as an end-to-end joint optimization problem of trace quotient maximization. The cost function is well-defined on the product of the CNN parameters space, the Stiefel manifold, the Oblique manifold, and the Grassmann manifold. By using the explicit gradient delivery, the cost function is optimized via a geometrical stochastic gradient descent (SGD) algorithm along with the chain rule and the backpropagation. The experimental results show that the proposed method can achieve a highly competitive performance with the state-of-the-art (SOTA) image classification, object categorization, and face recognition methods, under both supervised and semi-supervised settings. The code is available at https://github.com/MVPR-Group/SparConvLow.

A Convolutional Neural Network with Sparse Representation

Deep Neural Network Acceleration with Sparse Prediction Layers

Remote Sensing Scene Classification Using Sparse Representation-Based Framework with Deep Feature Fusion

LST-Net: Learning a Convolutional Neural Network with a Learnable Sparse Transform

A Pre-defined Sparse Kernel Based Convolution for Deep CNNs

SparseTrain: Exploiting Dataflow Sparsity for Efficient Convolutional Neural Networks Training

Fusing Deep Local And Global Features For Remote Sensing Image Scene Classification

Sparse Deep Transfer Learning for Convolutional Neural Network

SparseNet: A Sparse DenseNet for Image Classification

Pre-defined Sparsity for Low-Complexity Convolutional Neural Networks

A GPU-based high-performance optimization method of sparse convolutional neural networks

Integrating Convolution and Sparse Coding for Learning Low-Dimensional Discriminative Image Representations

Feature Sparsity in Convolutional Neural Networks for Scene Classification of Remote Sensing Image.

Hyperspectral Imagery Classification Using Sparse Representations of Convolutional Neural Network Features

Facial Expression Recognition Based on Convolutional Neural Network with Sparse Representation

Improvement of Pose Recognition by Sparse Regularized Convolutional Neural Network.

Convergence Analysis for Deep Sparse Coding via Convolutional Neural Networks

Revisiting Sparse Convolutional Model for Visual Recognition

Self-Representation Convolutional Neural Networks

Prune the Convolutional Neural Networks with Sparse Shrink

Learning Sparse Features with Lightweight ScatterNet for Small Sample Training