Table of Contents
This section features a number of tutorials illustrating some of the algorithms implemented in VLFeat, roughly divided into visual features such as SIFT and Fisher vectors and statistical methods, such as K-means, GMMs, KDTrees, and SVMs.
Visual features
Covariant detectors. An introduction to computing co-variant features like Harris-Affine.
Histogram of Oriented Gradients (HOG). Getting started with this ubiquitous representation for object recognition and detection.
Scale Invariant Feature Transform (SIFT). An introduction to SIFT keypoint and descriptor extraction and matching.
Dense SIFT (DSIFT) and PHOW. Extracting dense SIFT features for image classification.
Local Intensity Order Pattern (LIOP). Getting started with the LIOP descriptor as an alternative to SIFT in keypoint matching.
Maximally Stable Extremal Regions (MSER). Extracting MSERs from an image as an alternative covariant feature detector.
Image distance transform. Compute the image distance transform for fast part models and edge matching.
Fisher vector and VLAD encodings. Compute global image encodings by pooling local image features with Fisher vectors and VLAD.
Statistical methods
GMM. Learn Gaussian Mixture Models using the Expectation Maximization algorithm.
k-means. Cluster features with k-means.
Agglomerative Information Bottleneck (AIB). Cluster discrete data based on the mutual information between the data and class labels.
Quick shift. An introduction which shows how to create superpixels using this quick mode seeking method.
SLIC. An introduction to SLIC supoerpixels.
Support Vector Machine (SVM). Learn a binary classifier and check its convergence by plotting various statistical information.
Forests of kd-trees. Approximate nearest neighbor queries in high dimensions using an optimized forest of kd-trees.
Plotting functions for rank evaluation. Learn how to plot ROC, DET, and precision-recall curves.
MATLAB Utilities. A list of useful MATLAB functions bundled with VLFeat.
Obsolete tutorials
Integer optimized k-means (IKM). VLFeat integeger-otpimized k-means implementation (obsolete).
Hierarchical k-means (HIKM). Create a fast k-means tree for integer data (obsolete).