2012/02/28 by Arthur Szlam, Karol Gregor, Szlam, Arthur +3
Computer Science · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Image Retrieval and Classification Techniques #Machine Learning and Algorithms
paper · pdf · doi:10.48550/arxiv.1202.6384
openalex publication_date 2012/02/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We describe a method for fast approximation of sparse coding. The input space\nis subdivided by a binary decision tree, and we simultaneously learn a\ndictionary and assignment of allowed dictionary elements for each leaf of the\ntree. We store a lookup table with the assignments and the pseudoinverses for\neach node, allowing for very fast inference. We give an algorithm for learning\nthe tree, the dictionary and the dictionary element assignment, and In the\nprocess of describing this algorithm, we discuss the more general problem of\nlearning the groups in group structured sparse modelling. We show that our\nmethod creates good sparse representations by using it in the object\nrecognition framework of citelazebnik06,yang-cvpr-09. Implementing our own\nfast version of the SIFT descriptor the whole system runs at 20 frames per\nsecond on 321 \× 481 sized images on a laptop with a quad-core cpu, while\nsacrificing very little accuracy on the Caltech 101 and 15 scenes benchmarks.\n