一种用于大规模图像分类的可扩展学习算法

摘要

随着大规模图像分类数据集的出现，设计一种可扩展的、高效的多类分类算法成为目前一个重要的挑战。基于迹范数正则惩罚函数，提出了一种新的大规模多类图像分类的可扩展学习算法。把具有挑战性的非光滑优化问题重构为一个带l1正则惩罚的无穷维优化问题，进而设计了一个简单而有效的加速坐标下降算法。展示了如何在量化的密集视觉特征的压缩域中进行高效的矩阵计算，该压缩域有100000个例子，1000多维特征和100多类图片。在图像网的子集“Fungeus”，“Ungulate”和“Vehicles”上的实验结果表明，提出方法的性能明显优于目前最先进的16高斯Fisher向量方法。%With the advent of larger image classification datasets, designing scalable and efficient multi-class classifica-tion algorithms is now an important challenge. It introduces a new scalable learning algorithm for large-scale multi-class image classification, based on the trace-norm regularization penalty. Reframing the challenging non-smooth optimization problem into a surrogate infinite-dimensional optimization problem with a regular l1 regularization penalty, it proposes a simple and provably efficient accelerated coordinate descent algorithm. Furthermore, it shows how to perform efficient matrix computations in the compressed domain for quantized dense visual features, scaling up to 100000 examples, 1000-dimensional features, and 100 categories. Promising experimental results on the“Fungus”,“Ungulate”, and“Vehicles”subsets of ImageNet are presented, it shows that the approach performs significantly better than state-of-the-art approaches for Fisher vectors with 16 Gaussians.