Skip to content

Instantly share code, notes, and snippets.

@g2384

g2384/HoughForest.m

Created April 15, 2016 12:25
Show Gist options
  • Save g2384/4878b216f3437478e986682f6d6ccdf8 to your computer and use it in GitHub Desktop.
Save g2384/4878b216f3437478e986682f6d6ccdf8 to your computer and use it in GitHub Desktop.
Download ZIP
Raw
HoughForest.m
ntrees = 2; % number of trees
ratios = 1; %ratio
global min_sample;
min_sample = 2; % minimum number of samples/pixels
global max_depth;
max_depth = 15; % maximum depth
global num_sample;
num_sample = 10; % total number of samples/pixels
global num_image;
num_image = 10; % total number of samples/pixels
global max_w;
max_w = 100; % image width
global max_h;
max_h = 100; % image height
global max_c;
max_c = 2; % patch size
global num_leaf;
num_leaf = 0;
model = cell(1,ntrees); % forest
samples = [1];
%model = train(0,0,0,0);
%%
% function train(depth_part,label_part,coordinate,depth_whole)
for i = 1:ntrees
tree = [];
%%
% function growTree()
for l = 1:num_sample
fprintf('hello\n')
tree = grow(samples, 1, 1, tree);
end
%------------------
model{i} = tree;
end
%------------------
%%
#############################################
function tree = grow(samples, node, depth, tree)
global min_sample
global max_depth
global num_leaf
size_samples = size(samples)
if(depth < max_depth)
measure_mode = 1; % - = classification, 1 = regression
if(depth < max_depth - 2)
measure_mode = round(rand(1),0);
fprintf('MeasureMode %d, depth %d',measure_mode, depth);
test = [];
A = [];
B = [];
t = 0;
t, A, B, test = optimizeTest(samples, size_samples, measure_mode);
if(t == 1)
tmp = test;
countA = 0, countB = 0;
tree(node) = test;
for l = 0:size_samples
fprintf('Final Split %d: A - %d, B - %d', l, size(A), size(B));
end
if size(A)>min_sample
tree = grow(A, 2*node+1,depth+1, tree);
else
tree = makeLeaf(A, 2*node+1, tree,num_leaf);
num_leaf =num_leaf+ 1;
end
if size(B)>min_sample
tree = grow(B, 2*node+2,depth+1, tree);
else
tree = makeLeaf(B, 2*node+2, tree,num_leaf);
num_leaf =num_leaf+ 1;
end
else
tree = makeLeaf(B, 2*node+2, tree,num_leaf);
num_leaf =num_leaf+ 1;
end
else
tree = makeLeaf(B, 2*node+2, tree,num_leaf);
num_leaf =num_leaf+ 1;
end
end
end
function [t, A, B, test] = optimizeTest(samples, size_samples, measure_mode)
t = 0;
bestDist = 9999;
test = [];
A = [];
B = [];
global max_w;
global max_h;
global max_c;
global num_image;
for i = 1:num_image
%%
% generateTest()
test = [round(rand()*max_w), round(rand()*max_h),round(rand()*max_w), round(rand()*max_h), round(rand()*max_c)];
%%
% evaluateTest()
valSet_value = []
vmin = 999;
vmax = 0;
for l = 1:size_samples
value = samples(max_w*test(1)+test(2))-samples(max_w*test(3)+test(4));
valSet_value = [valSet_value, value];
if vmin > value
vmin = value;
elseif vmax < value
vmax = value;
end
end
d = vmax - vmin;
if d>0
for j = 1:10
tr = round(rand(d)) + vmin;
%%
% split()
t = valSet_value<tr;
[mx,ix] = max(valSet_value(t));
A = samples(1:ix);
B = samples(ix+1:size_samples);
A_size = size(A)
B_size = size(B)
if(A_size > 0 && B_size > 0)
%%
% measureSet()
%if measure_mode == 0
% info gain
cdist = histc(A, unique(A)) + 1;
cdist = cdist/sum(cdist);
cdist = cdist .* log(cdist);
n_A = -sum(cdist);
cdist = histc(B, unique(B)) + 1;
cdist = cdist/sum(cdist);
cdist = cdist .* log(cdist);
n_B = -sum(cdist);
tmpDist = (A_size*n_A+B_size*n_B)/(A_size+B_size);
%else
%end
if tmpDist > bestDist
f = 1;
bestDist = tmpDist;
test = [test, tr];
end
end
end
end
end
end
function tree = makeLeaf(set, node, tree,num_leaf)
tree(node) = num_leaf;
end
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment