Visual Geometry Group 资源（vgg）

1.Phog 代码：http://www.robots.ox.ac.uk/~vgg/research/caltech/phog.html

其它链接：http://www.robots.ox.ac.uk/~vgg/software/

Homogeneous kernel map for fast non-linear classification with additive kernels

The homogeneous kernel map allows large scale training of non-linear SVMs based on additive kernels such as Chi2, intersection, and Jensen-Shannon. The map transforms the data into a compact linear representation which reproduces the desired kernel to a very good level of approximation. This representation enables then to use very fast linear SVM solvers.

User interface for interactive image segmentation

A matlab implementation of a user interface for interactive segmentation. Implements the star-convexity algorithms described in Gulshan et al. in CVPR2010 and other commonly used interactive segmentation methods.

FASTANN and FASTCLUSTER for approximate k-means (AKM)

A distributed implementation of the approximate k-means (AKM) algorithm presented in Philbin et al. at CVPR 2007. The software consists of two libraries: i) FASTANN, a library for fast, approximate nearest neighbours. ii) FASTCLUSTER, an MPI-distributed library for doing exact and approximate k-means.

Descriptor Learning Using Convex Optimisation

Code and learnt models for feature descriptor computation and evaluation, as described in Simonyan et al., ECCV 2012.

2D articulated human pose estimation

The software for articulated human pose estimation in still images is designed to operate in uncontrolled images with difficult illumination conditions and cluttered backgrounds. The only assumption the algorithm makes is that people are upright (i.e. their head is above their torso) and they are seen approximately from a frontal viewpoint.

Hand detection using multiple proposals

The code for hand detection in static images implementing the method described in Mittal et al. at BMVC 2011.

Efficient structured output SVM ranking

The reference implementation of the structured output ranking algorithm with linear time constraint generation method proposed in Mittal et al. at ECCV 2012.

Encoding Methods Evaluation Toolkit

MATLAB code for evaluation of different bag of visual words encoding schemes over standard image classification test datasets.

Multiple Kernel Learning for Image Classification

The VGG MKL classifier is an implementation of multiple kernel learning for image classification. It bundles powerful image descriptors (spatial pyramids of geometric blur, fast dense SIFT, and self-similarity features) and achieves state-of-the-art performance on Caltech-101. In MATLAB/C, compatible with Linux 32/64, Windows 32/64, and Mac OS X.

Upper-body Detector

The upper-body detector software pages provide download links for software designed to detect the region between the top of the head and the upper half of the torso. Example results (images and video) and performance evaluations are included.

Self-Similarity Descriptor

Implementation of the Self-Similarity Descriptor by Varun Gulshan, based on the paper Matching Local Self-Similarities across Images and Videos, by Eli Shechtman and Michal Irani at CVPR '07.

Multi-frame Image Super-resolution

The Super-resolution code page provides a basic suite of Matlab/C-Mex functions for computing ML and MAP super-resolution image estimates, including documentation and a demo m-file.

Pyramid Histogram of Oriented Gradients

Code for computing the Pyramid Histogram of Oriented Gradients (PHOG) descriptor over a Region Of Interest (ROI) is provided on the phog page, which is part of theImage Classification section of the research pages.

Affine Covariant Features

The following items link into the Affine Covariant Features section of the VGG'sresearch pages.

• Region detectors - Linux binaries for detecting affine covariant regions.
• Region descriptors - Linux binaries for computing region descriptors.
• Detectors evaluation - Matlab files to compute the repeatability.
• Descriptors evaluation - Matlab files to compute the matching score.

 

Affine Normalized Regions

input

output

Linux binary to compute affine normalized regions around interest points. With thanks to Frederik Schaffalitzky, VGG, Univ of Oxford.

 

Probabilistic Latent Semantic Analysis (pLSA)

• pLSA Matlab demo code (by Josef Sivic, VGG, Univ. of Oxford)
• An extended version of this pLSA code was included in the ICCV 2005 short course Recognizing and Learning Object Categories by Li Fei Fei, Rob Fergus and Antonio Torralba and can be downloaded from the short coursewebpage.
• Another implementation of the pLSA model (by Peter Gehler)