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.

Learning a Discriminative Projection and Representation for Image Classification

Learnable Subspace Orthogonal Projection for Semi-supervised Image Classification

Learning discriminative representation for image classification

Graph Learning in Low Dimensional Space for Graph Convolutional Networks

A-Optimal Projection for Image Representation.

Local Feature Discriminant Projection

Structure Preserving Projections Learning Via Low-Rank Embedding for Image Classification

Fast Low Rank Representation Based Spatial Pyramid Matching for Image Classification.

Jointly Learning Non-negative Projection and Dictionary with Discriminative Graph Constraints for Classification

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

Robust L1-Norm Matrixed Locality Preserving Projection For Discriminative Subspace Learning

Laplacian Sparse Dictionary Learning for Image Classification Based on Sparse Representation

Fuzzy Discriminative Block Representation Learning for Image Feature Extraction

Discriminative Regression With Latent Label Learning for Image Classification

Structurally Incoherent Low-Rank 2DLPP for Image Classification

Low-rank Representation Based Discriminative Projection for Robust Feature Extraction

Low-rank discriminative least squares regression for image classification

Sparsity Induced Locality Preserving Projection Approaches for Dimensionality Reduction

Regularisation Constrained Denoising Discriminant Least Squares Regression for Image Classification

Random Projections for Classification: A Recovery Approach

Double Constrained Discriminative Least Squares Regression for Image Classification