image thumbnail

Fast GMM and Fisher Vectors

by

 

27 Sep 2012 (Updated )

Fast GMM (diagonal covariances only) with Kmeans initialization and Fisher Vectors

[GX,GY]=gen_dgauss(sigma)
function [GX,GY]=gen_dgauss(sigma)

% laplacian of size sigma
%f_wid = 4 * floor(sigma);
%G = normpdf(-f_wid:f_wid,0,sigma);
%G = G' * G;

if(nargin < 1)
    
    sigma = 1;
    
end

G       = gen_gauss(sigma);
[GX,GY] = gradient(G); 

GX      = GX * 2 ./ sum(sum(abs(GX)));
GY      = GY * 2 ./ sum(sum(abs(GY)));

Contact us