Fast approximations to structured sparse coding and applications to object classification

TL;DR

This paper introduces a fast approximation method for structured sparse coding using a decision tree, enabling real-time object classification with minimal accuracy loss.

Contribution

It presents a novel decision tree-based approach for rapid sparse coding, including algorithms for learning the tree, dictionary, and group assignments, improving inference speed.

Findings

Achieves 20 frames per second on standard images

Maintains high accuracy on Caltech 101 and 15 scenes benchmarks

Enables real-time object recognition with minimal accuracy compromise

Abstract

We describe a method for fast approximation of sparse coding. The input space is subdivided by a binary decision tree, and we simultaneously learn a dictionary and assignment of allowed dictionary elements for each leaf of the tree. We store a lookup table with the assignments and the pseudoinverses for each node, allowing for very fast inference. We give an algorithm for learning the tree, the dictionary and the dictionary element assignment, and In the process of describing this algorithm, we discuss the more general problem of learning the groups in group structured sparse modelling. We show that our method creates good sparse representations by using it in the object recognition framework of \cite{lazebnik06,yang-cvpr-09}. Implementing our own fast version of the SIFT descriptor the whole system runs at 20 frames per second on $321 \times 481$ sized images on a laptop with a quad-core cpu, while sacrificing very little accuracy on the Caltech 101 and 15 scenes benchmarks.