plateform stringclasses 1
value | repo_name stringlengths 13 113 | name stringlengths 3 74 | ext stringclasses 1
value | path stringlengths 12 229 | size int64 23 843k | source_encoding stringclasses 9
values | md5 stringlengths 32 32 | text stringlengths 23 843k |
|---|---|---|---|---|---|---|---|---|
github | salimoha/delta-Dogs-Lambda-master | print2eps.m | .m | delta-Dogs-Lambda-master/FilesToPath/tools/print2eps.m | 6,118 | utf_8 | d51322b49255fd17900b2a402b760e8e | %PRINT2EPS Prints figures to eps with improved line styles
%
% Examples:
% print2eps filename
% print2eps(filename, fig_handle)
% print2eps(filename, fig_handle, options)
%
% This function saves a figure as an eps file, with two improvements over
% MATLAB's print command. First, it improves the line style, makin... |
github | salimoha/delta-Dogs-Lambda-master | convex_bnd_project.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/convex_bnd_project.m | 926 | utf_8 | a93cb369387a70688bb4b3796f07f736 | function [xp,xc,R2,ii] = convex_bnd_project(x,xi,tri)
% Find the convex boundary projection of the point x
global n m cons
out=0;
for ii=1:size(tri,1)
if (min(tri(ii,:))<n+2)
[xc,R2]=circhyp(xi(:,tri(ii,:)), n);
if norm(x-xc)< sqrt(R2)
out=1;
... |
github | salimoha/delta-Dogs-Lambda-master | quad_inter_paramterization.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/quad_inter_paramterization.m | 1,441 | utf_8 | f3f86397a96ad96dff376d924b7499ed | function inter_par=quad_inter_paramterization(xi1,yi1)
global n y0 xi yi w
%keyboard
[tt,ind]=min(yi1);
xi=xi1; yi=yi1;
xmin=xi(:,ind); xi(:,ind)=[];
xi=xi-repmat(xmin,1,size(xi,2));
ymin=tt; yi(ind)=[];
yi=yi-ymin;
%keyboard
f=zeros(n,1); A=eye(n);
x0=[f;reshape(A,n^2,1)];
%keyboard
% weights of the interpolation
Tol=... |
github | salimoha/delta-Dogs-Lambda-master | regressionparametarization.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/regressionparametarization.m | 1,539 | utf_8 | a1fda62b2998e62d648262da3afb5a66 | function [inter_par,yp]= regressionparametarization(xi,yi,sigma,inter_method)
n=size(xi,1);
N=size(xi,2);
%while 1
if inter_method==1
% keyboard
A = zeros(N,N);
% calculate regular A matrix for polyharmonic spline
for ii = 1 : 1 : N
for jj = 1 : 1 : N
A(ii,jj) = ((xi(:,ii) - xi(:,jj))' * (xi(:,ii)... |
github | salimoha/delta-Dogs-Lambda-master | Linear_constrained_initilazation.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/Linear_constrained_initilazation.m | 958 | utf_8 | 7100f569f04d09e01bbfebe99dfa977b | function [xi,xic] = Linear_constrained_initilazation(Ain, bin)
% Initialization of the linearly_cosntrained problem
% Ain x \leq bin
global n kappa
kappa=1.5;
% Find an initial interior point for L
options=optimoptions('linprog','Display','off');
x1 = linprog([zeros(1,n) -1],[Ain ones(size(Ain,1),1); [zeros(2,n) [1... |
github | salimoha/delta-Dogs-Lambda-master | feasible_point_finder.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/feasible_point_finder.m | 988 | utf_8 | de1df4e3130efb26dcc2828244271283 | function [x]=feasible_point_finder(x,H,f,e)
% Calculate an initial feasible point for the feasible domain
% minimize -\sum log(-x'H_ix- f'x_i- e_i)
global m n
for i=1:m
e{i}=e{i}-0.1;
end
while 1
[M]=cost_fun(x,H,f,e);
if (M==0)
break
end
[dM]=grad_fun(x,H,f,e);
[d2M]=hess_fun(x,H,f,e);
d2M=d2M+0.01*eye(n);
p=... |
github | salimoha/delta-Dogs-Lambda-master | Adoptive_K_Search.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/Adoptive_K_Search.m | 1,693 | utf_8 | 3a33c3f3104c23ea4264fe5d280331c8 | function [x y cse]=Adoptive_K_Search(x,inter_par,xc,R2)
% Find the minimizer of the search function in a simplex
global n Ain bin
cc=0.01; rho=0.9; % parameters of backtracking
% Initialize the point in the simplex
%[xc,R2]=circhyp(xiT(:,tri(index,:)), n);
iter=1; cse=2;
y=cost(x,inter_par,xc, R2);
while iter<3
% Cal... |
github | salimoha/delta-Dogs-Lambda-master | tringulation_search_bound_constantK.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/tringulation_search_bound_constantK.m | 1,411 | utf_8 | 17915f235e824916ef37e0fc09351d11 | function [xm ym] = tringulation_search_bound_constantK(inter_par,xi,K,ind_min)
global n
tri=delaunayn(xi.');
%keyboard
% Search over the simplices
for ii=1:size(tri,1)
[xc,R2]=circhyp(xi(:,tri(ii,:)), n);
x=xi(:,tri(ii,:))*ones(n+1,1)/(n+1);
Sc(ii)=interpolate_val(x,inter_par)-K*(R2-norm(... |
github | salimoha/delta-Dogs-Lambda-master | inter_min_conv.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/inter_min_conv.m | 1,354 | utf_8 | c6ac69b65500e9515fd2f03a15b40907 | function [x y]=inter_min_conv(x, inter_par)
%find the minimizer of the interpolating function starting with x
global n
rho=0.9; % parameters of backtracking
% start the search with method
mu=1; iter=1;
while iter<4
x_pre=x; iterr=1;
while iterr<10
% Calculate the Newton direction
y=cost(x,inter_par,mu);
g=grad(x,... |
github | salimoha/delta-Dogs-Lambda-master | actual_scaling_lincon.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/actual_scaling_lincon.m | 884 | utf_8 | 153ad2429dded43bc72d09803deafc78 | function [A,B,s,r]=actual_scaling_lincon(A,B,ub,lb,a,b,n,neq)
% find the real scaling factor
A=[A; eye(n);-eye(n)];
B=[B; ub; -lb];
% Find all vertice of Ax \leq B,
% and the nonredundent constraints
[X,NRC]=vertexfind(A,B,a,b,n,neq);
s=min(X.').'; s1=max(X.').';
r=s1-s;
s=s(1:n-neq); r=r(1:n-neq);
A=A(NRC,:); B=B(N... |
github | salimoha/delta-Dogs-Lambda-master | Initial_point_finder.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/Initial_point_finder.m | 1,102 | utf_8 | 4644846ce21e52670a25f618daca6fa3 | function [x]=Initial_point_finder(x,H,f,e)
% calculate an initial interior point for the feasible domain
% minimize -\sum log(-x'H_ix- f'x_i- e_i)
global m n
while 1
[M]=cost_fun(x,H,f,e);
[dM]=grad_fun(x,H,f,e);
[d2M]=hess_fun(x,H,f,e);
p=-d2M\dM;
%line search method
for j=1:m
alpha(j) = quadratic_prolongation(x,... |
github | salimoha/delta-Dogs-Lambda-master | interpolateparametarization.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/interpolateparametarization.m | 11,185 | utf_8 | 2a942628fb0311d2986712e82e7f3298 | function inter_par= interpolateparametarization(xi1,yi1,inter_method,interpolate_index)
global xi yi y0 w
xi=xi1; yi=yi1;
n=size(xi,1);
% keyboard
% polyharmonic spline interpolation
if inter_method==1
N = size(xi,2); A = zeros(N,N);
for ii = 1 : 1 : N
for jj = 1 : 1 : N
A(ii,jj) = ((xi(:,ii) - xi(:,jj... |
github | salimoha/delta-Dogs-Lambda-master | Adoptive_K_conv.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/Adoptive_K_conv.m | 2,135 | utf_8 | 432e4876f96e9c8f9a2164804f794fdb | function [x y cse]=Adoptive_K_conv(x,inter_par,xc,R2)
% Find the minimizer of the search function in a simplex
rho=0.9; % parameters of backtracking
% Initialize the point in the simplex
%[xc,R2]=circhyp(xiT(:,tri(index,:)), n);
iter=1; mu=1; cse=2;
x_pre=x;
while iter<10
iterr=1;
while iterr<10
y=cost(x,inter_par,xc... |
github | salimoha/delta-Dogs-Lambda-master | lin_convex_bnd_project.m | .m | delta-Dogs-Lambda-master/FilesToPath/Search/lin_convex_bnd_project.m | 610 | utf_8 | ce3e89635467b7fc8a38db30ab71e359 | function [xp,xc,R2,ii] = lin_convex_bnd_project(x,xi,tri)
% Find the convex boundary projection of the point x
global n m Ain bin
lincon=1;
out=0;
for ii=1:size(tri,1)
if (min(tri(ii,:))<n+2)
[xc,R2]=circhyp(xi(:,tri(ii,:)), n);
if norm(x-xc)< sqrt(R2)
ou... |
github | cbaldassano/Event-Segmentation-master | forward_backward_log.m | .m | Event-Segmentation-master/forward_backward_log.m | 2,102 | utf_8 | 9c81dc43386d9b69a3f9fbb875905a5e | function [loggamma, LL] = forward_backward_log(logprob, Pi, EndPi, P)
% Computes log p(event at time t = k), and log likelihood of fit
% Runs modified forward-backward algorithm that allows for ending event
% constraing (EndPi)
% Inputs:
% logprob: time by event matrix of log p(response t | event k)
% Pi, End... |
github | cbaldassano/Event-Segmentation-master | example.m | .m | Event-Segmentation-master/example.m | 3,296 | utf_8 | e4f265baae048f1228055d6409240b4c | function example()
% Example of event segmentation and finding corresponding events
% Parameters for creating small simulated datasets
V = 10;
K = 10;
T = 500;
T2 = 300;
% Generate the first dataset
rng(1);
eventMeans = randn(V,K);
eventLabels = generate_event_labels(T, K, 0.1);
simulData = generate_data... |
github | stoneyang/caffe_ristretto-master | classification_demo.m | .m | caffe_ristretto-master/matlab/demo/classification_demo.m | 5,412 | utf_8 | 8f46deabe6cde287c4759f3bc8b7f819 | function [scores, maxlabel] = classification_demo(im, use_gpu)
% [scores, maxlabel] = classification_demo(im, use_gpu)
%
% Image classification demo using BVLC CaffeNet.
%
% IMPORTANT: before you run this demo, you should download BVLC CaffeNet
% from Model Zoo (http://caffe.berkeleyvision.org/model_zoo.html)
%
% *****... |
github | jjcharles/personalized_pose-master | train_rgb_model.m | .m | personalized_pose-master/+evaluator/train_rgb_model.m | 8,273 | utf_8 | 07f28ed0d7c614b174e04b73a49dc42b | %get lower arm likelihood values
%leaves a hold out validation set to tune confidence values
function [left_shape,right_shape] = train_rgb_model(opts,filename,folder,detections,fieldname,model_scale)
visualise = false;
frameids = detections.(fieldname).frameids;
locs = detections.(fieldname).locs;
total_frame... |
github | jjcharles/personalized_pose-master | best_detections.m | .m | personalized_pose-master/+evaluator/best_detections.m | 2,296 | utf_8 | 98064cd152f918bf8b0b5b519694a667 | %function to evaluate the joint detections and return those which we
%believe to have high confidence
%takes as input the detections struct, outputs a detections struct
%with added fields for filtered detections
function detections = best_detections(detections)
%for each joint find best detections by thresho... |
github | jjcharles/personalized_pose-master | sample_detections.m | .m | personalized_pose-master/+evaluator/sample_detections.m | 1,551 | utf_8 | b4100b4c44540ca01f46994b14c31c4c | %resample detections so that there is a maximal coverage of locations
%across each object/body_part type
%normalises to head position
function detections = sample_detections(detections,field_name,patch_width,num_samples_per_part)
num_parts = size(detections.(field_name).locs,2);
locs = [];
fra... |
github | jjcharles/personalized_pose-master | apply_hogrgb_model.m | .m | personalized_pose-master/+evaluator/apply_hogrgb_model.m | 6,105 | utf_8 | 290d47d47515a62ef1509d4d7b492a8f | %get lower arm likelihood values by using both a hog and an rgb model
function [detections, removed, fixed] = apply_hogrgb_model(filename,folder,detections,fieldname,hogmodel, rgbmodel,model_scale)
%detections - filtered detections
%removed - ids of where old detections were removed
visualise = false;
frameids... |
github | jjcharles/personalized_pose-master | remove_invalid_poses.m | .m | personalized_pose-master/+evaluator/remove_invalid_poses.m | 6,483 | utf_8 | d6b016f4786a28cce3fcaabb910f57a4 | %function to identify frames where the pose estimates are not valid, ie.
%they have limb lengths greater than normal
function detections = remove_invalid_poses(detections,field_name,current_detections)
shoulder_to_shoulder = sqrt(sum( (current_detections.manual.locs(:,[6],:)-current_detections.manual.locs(:,[... |
github | jjcharles/personalized_pose-master | foreground_background_eval.m | .m | personalized_pose-master/+evaluator/foreground_background_eval.m | 3,375 | utf_8 | 037cac9d0562700bb9bce1694220f27e | %function to remove joint detections based on foreground background
%classification
%allows input of seperate training and testing data
function detections_test = foreground_background_eval(opts,filename,detections_train,detections_test,fieldname)
frameids = detections_train.training.frameids;
locs = detections_... |
github | jjcharles/personalized_pose-master | train_hog_model.m | .m | personalized_pose-master/+evaluator/train_hog_model.m | 9,367 | utf_8 | e292a7c8c7c94dd36352e318ccdc4c5b | %get lower arm likelihood values
%uses jittered positive examples
%trains the svms with a threshold by validating on hold out data
function [left_shape,right_shape] = train_hog_model(opts,filename,folder,detections,fieldname,model_scale)
visualise = false;
frameids = detections.(fieldname).frameids;
locs = de... |
github | jjcharles/personalized_pose-master | get_hogrgb_models.m | .m | personalized_pose-master/+evaluator/get_hogrgb_models.m | 1,188 | utf_8 | 87551250d7c9fc8148df256a67efd2a1 | %function which trains an svm evaluator for both hog and rgb features
%it returns two different models one for hog and one for rgb
%uses a hold out set of validation data to tune the svm thresholds
function [hogmodel, rgbmodel] = get_hogrgb_models(opts,filename,folder,detections,fieldname,modelscale)
%subsample... |
github | jjcharles/personalized_pose-master | get_training_patches.m | .m | personalized_pose-master/+occlusion/get_training_patches.m | 3,476 | utf_8 | ed390106f7ed601188b96818b6b6e0ab | %returns both hog and rgb patches with rotation augmentation
function [hogpatches,rgbpatches, labels] = get_training_patches(opts,videofilename,frameids,locs)
%GET_TRAINING_PATCHES_HOG gets patches of image for training an occlusion
%detector
%get size of hog patches
dummypatch = single(round(rand(opts.occlusi... |
github | jjcharles/personalized_pose-master | apply_forest_fast.m | .m | personalized_pose-master/+oforest/apply_forest_fast.m | 14,275 | utf_8 | 995485fe08cabb881e799022258ab021 | %function to apply forest to a frame
%interlaces the application of trees on every other pixel, this gives a
%theoretical speed increase of x2
function [locs,dist,conf,leafids] = apply_forest_fast(omodel,frame,bbox,islocs_needed)
%check for correct padding
omodel.opts.model.windowwidth = max(omodel.opts.m... |
github | jjcharles/personalized_pose-master | train_forest_cluster.m | .m | personalized_pose-master/+oforest/train_forest_cluster.m | 395 | utf_8 | eea08e96ab3a71320219a3fc6a0874d5 | %TRAIN_FOREST - trains a part detector classifier for object locations
%implementation to run on cluster
function tree = train_forest_cluster(opts,filenames,frame_ids,locs,treeid)
fprintf('Training part detectors (%d images):...',numel(frame_ids));
rseed = treeid;
tree = oforest.master_node_fast(op... |
github | jjcharles/personalized_pose-master | load_images.m | .m | personalized_pose-master/+oforest/load_images.m | 4,421 | utf_8 | d478eabb96ae1524ffd7af7c5263de2d | %function to load training images into memory
function [images, labels, sample_idx,total_background, newpadding] = load_images(opts,filename_video, frame_ids, locs)
%images - the training images pluse a few background unlabelled images
%labels - this is actuall the body part locations
%sample_idx - the sample index... |
github | jjcharles/personalized_pose-master | apply_frame.m | .m | personalized_pose-master/+oforest/apply_frame.m | 1,328 | utf_8 | ce1b1cef62df72921480a69d85c9e17f | %function to apply the general forest to a frame
function [locs, dist, conf, leafids] = apply_frame(omodel,img,seg, patches, bbox, featuretype, testing_speed,filter_width)
%bbox is the bounding box where the forest is applied
if ~exist('filter_width','var')
filter_width = 4;
end
switch... |
github | jjcharles/personalized_pose-master | multiclass_thresh.m | .m | personalized_pose-master/+oforest/multiclass_thresh.m | 1,929 | utf_8 | 65dc0560d37695c231a07c7750228666 | %MULTICLASS_THRESH - find multiclass threshold value for tree
function [T, Gmax, window_index_left, window_index_right] = multiclass_thresh(opts,WI,data,feature,channel,func_type)
window_index_left = [];
window_index_right = [];
Gmax = -inf;
data_class = data.class(WI);
num_samples = siz... |
github | jjcharles/personalized_pose-master | augmentdata.m | .m | personalized_pose-master/+oforest/augmentdata.m | 1,646 | utf_8 | d310820ce59d0fa08f7572f202df84e0 | %augment the training data by adding in some rotations
function [images,labels] = augmentdata(input_images,input_labels,rotations)
images = repmat(uint8(0),[size(input_images,1),size(input_images,2),size(input_images,3),size(input_images,4)*(1+numel(rotations))]);
labels = zeros(size(input_labels,1),size(i... |
github | jjcharles/personalized_pose-master | apply_forest.m | .m | personalized_pose-master/+oforest/apply_forest.m | 2,363 | utf_8 | b909fa0bb0ff788632a861a74ee33117 | %function to apply forest to a frame
function [locs,dist,conf,leafids] = apply_forest(omodel,frame,bbox,islocs_needed,filter_width)
if ~exist('filter_width','var')
filter_width = 8;
end
if ~exist('islocs_needed','var')
islocs_needed = true;
end
%check for correct padd... |
github | jjcharles/personalized_pose-master | sample_windows.m | .m | personalized_pose-master/+oforest/sample_windows.m | 4,219 | utf_8 | 1daea87e95419d0fc4f7677a31e3a6f1 | %SAMPLE_WINDOWS - samples with windows from the loaded images
function points = sample_windows(opts,labels,total_background)
%initialise the points array
numbackgroundextras = 100;
patch = oforest.make_patch(opts.patchwidth);
total_points = 0;
for i = 1:size(labels,3)
numspots = 0;
... |
github | jjcharles/personalized_pose-master | get_smaller_sample.m | .m | personalized_pose-master/+oforest/get_smaller_sample.m | 1,282 | utf_8 | 5b1ce1be13d3d11219d50b5bb0564c82 | %function to sample a uniform selection of windows across all classes
function [WI_small,winid,norm_class_dist] = get_smaller_sample(WI,data,class_dist,maxsample)
num_diff_classes = numel(class_dist);
N = sum(class_dist);
if N==0
class_dist = histc(data.class(WI),1:num_diff_classes);
... |
github | jjcharles/personalized_pose-master | get_feature.m | .m | personalized_pose-master/+oforest/get_feature.m | 1,455 | utf_8 | b92258ea9b0e8dfcad594254d2128351 | %function to extract feature from images given data points and offset
function feature = get_feature(opts,WI,data,images,channel,func_type,offset)
func = oforest.func_pointer(func_type);
data_img_index = double(data.img_index(WI));
data_x = double(data.x(WI));
data_y = double(data.y(WI));
... |
github | jjcharles/personalized_pose-master | master_node_fast.m | .m | personalized_pose-master/+oforest/master_node_fast.m | 6,935 | utf_8 | a1aed67c4fc91935bfa814288698b716 | %MASTER_NODE - builds multiclass decision tree.
function [tree,model_opts] = master_node_fast(opts, filename_video, frame_ids, locs, random_seed)
model_opts = opts.model;
model_opts.bbox = opts.bbox;
model_opts.imgwidth = opts.bbox(3) + 2*model_opts.padding;
model_opts.imgheight =opts.bbox(4) + 2*m... |
github | jjcharles/personalized_pose-master | make_patch.m | .m | personalized_pose-master/+oforest/make_patch.m | 172 | utf_8 | f6792896dd107cb9fbd42233a689279b | %make a canonical joint patch
function patch = make_patch(width)
disc = strel('disk',width,0);
[r,c] = find(disc.getnhood);
patch = [c, r] - width - 1;
end |
github | jjcharles/personalized_pose-master | sample_node_data.m | .m | personalized_pose-master/+oforest/sample_node_data.m | 2,132 | utf_8 | 51c16741692b7e2d9477ed9da9bb0460 | %SAMPLE_NODE_DATA - samples test functions and features for node
function [feature, offset] = sample_node_data(opts,WI,data,images,channel,func_type,currentdepth)
func = oforest.func_pointer(func_type);
%decide what window width to use
if exist('currentdepth','var')
if isfield(opts,'wi... |
github | jjcharles/personalized_pose-master | apply_video.m | .m | personalized_pose-master/+oforest/apply_video.m | 1,241 | utf_8 | aa154c53b2ac949797fd8601dcd22348 | %function to apply partdetector to video
function [locs,confout,dist] = apply_video(omodel,videofilename,frameids,model_scale)
orig_bbox = omodel.opts.bbox;
if ~exist('model_scale','var')
model_scale = 1;
else
omodel = oforest.scale_model(omodel,model_scale);
end
vidobj = VideoReader(videofilena... |
github | jjcharles/personalized_pose-master | load_forest_from_folder.m | .m | personalized_pose-master/+oforest/load_forest_from_folder.m | 1,195 | utf_8 | 041b63dbcceae0b2a11a18684059ee1d | %function to load oforest from folder containing trees
function model = load_forest_from_folder(folder)
folder(folder=='\') = '/';
if folder(end)~='/'; folder = [folder '/']; end
forestfile = sprintf('%sforest.mat',folder);
model = struct();
if ~exist(forestfile,'file');
%get... |
github | jjcharles/personalized_pose-master | func_pointer.m | .m | personalized_pose-master/+oforest/func_pointer.m | 647 | utf_8 | f080d16ad700645bb3b2a1a589eeeab3 | %provides a pointer to a function indexed by a number
function fp = func_pointer(num)
switch num
case 1
fp = @unary;
case 2
fp = @binary1;
case 3
fp = @binary2;
case 4
fp = @binary3;
otherwise
fp = @unary... |
github | jjcharles/personalized_pose-master | tree2mat.m | .m | personalized_pose-master/+oforest/tree2mat.m | 1,055 | utf_8 | aae794e8eb4b11bdd89498c348ea0bc2 | %function to convert a ctree into matrix form
function mat_tree = tree2mat(tree)
numclasses = numel(tree(end).distribution);
%preallocate memory
mat_tree = zeros(11+numclasses,numel(tree));
for n = 1:numel(tree)
if ~isempty(tree(n).left); mat_tree(1,n) = (tree(n).left-1)*19... |
github | jjcharles/personalized_pose-master | scale_model.m | .m | personalized_pose-master/+oforest/scale_model.m | 336 | utf_8 | ee4dbd04285fe725216c8bb83760708a | %function to scale test functions of model to work with different sized
%images
function model = scale_model(model,scale)
for f =1:numel(model.forestmat)
model.forestmat{f}(5:8,:) = floor(model.forestmat{f}(5:8,:)*scale);
end
model.opts.bbox(:) = floor(model.opts.bbox(:)*scale);
model.opts.bbox(model.opt... |
github | jjcharles/personalized_pose-master | get_face_torso_patches.m | .m | personalized_pose-master/+prep/get_face_torso_patches.m | 3,171 | utf_8 | 295caa9699778e06e91e70b235a9678a | %function to get the face and torso patches
function videoToColor = get_face_torso_patches(opts,videofilename,model,filename,frameids)
vidobj = VideoReader(videofilename);
if ~exist('frameids','var');
frameids = 1:20:vidobj.NumberOfFrames;
end
videoToColor = [];
frameids = fra... |
github | jjcharles/personalized_pose-master | temporal_refinement.m | .m | personalized_pose-master/+flow/temporal_refinement.m | 8,726 | utf_8 | 0b8b056914ba3f98f5ac9ac3b1099bb0 | %function to refine poses by comparing them against propergated
%neighbouring frames
function [refinedlocs,refinedframes,alllocs,allframeids,fromframeid,allflowq] = temporal_refinement(opts,filename,input_locs,input_frameids)
visualise = false;
refinedlocs = [];
refinedframes = [];
%open f... |
github | jjcharles/personalized_pose-master | cluster_refinement_highres.m | .m | personalized_pose-master/+flow/cluster_refinement_highres.m | 4,475 | utf_8 | 6ceb20d7ba4d763549ea4eb65ebef196 | %function to refine pose using dense siftflow WITHIN the cluster
function [poses, from_frame_idx, flowquality, initial_frameids] = cluster_refinement_highres(part_clusters,initial_frameids,videofilename,apply_to_locs)
%part_clusters are clustererd patches around body part detections
%initial_frameids = initial sampl... |
github | jjcharles/personalized_pose-master | align_locs.m | .m | personalized_pose-master/+flow/align_locs.m | 2,241 | utf_8 | a61b6bf72a8c7a75919963ec354d8390 | %function to propergate a single point through a set of matching frames
function [locarray, energies, from_patch_idx] = align_locs(patches,initial_patch)
%locarray is the array of joint locations relative to the patch centre
%energies is the flow energy for each patch
%from_frame_idx holds the patchidx from where t... |
github | jjcharles/personalized_pose-master | combine_proposals.m | .m | personalized_pose-master/+flow/combine_proposals.m | 9,637 | utf_8 | 58e1fd6fa49ba37e1a9541a4a0f0267b | %function to combine proposals and formulate the statistics per frame
%works using detection struct as input
%provides a contribution field with all stats
function detections = combine_proposals(detections,videofilename)
visualise = true;
if visualise && exist('videofilename','var')
vidobj... |
github | jjcharles/personalized_pose-master | get_backwards_flow.m | .m | personalized_pose-master/+flow/get_backwards_flow.m | 1,255 | utf_8 | c08a806c4aad25c39f5f31184c7b261d | %function to produce flow backwards from flow going fowards using iterative fixed
%point algorithm: flowback_nplus(x) = -flowforward(x+flowback_n(x));
%initialise flowback_zero = -flowforward;
function reverse_flow = get_backwards_flow(flow)
reverse_flow = zeros(size(flow));
[M,N,~,total_frames] = size(f... |
github | jjcharles/personalized_pose-master | compress_flow_minmax.m | .m | personalized_pose-master/+flow/compress_flow_minmax.m | 498 | utf_8 | 85252f13aafbf47d9a28df6121c7665d | %function to read in flow file and compress
function [flow_small, minmax] = compress_flow_minmax(filename)
flow = flow.readFlowFile(filename);
minmax = [max(max(flow(:,:,1))), min(min(flow(:,:,1)));... % channel 1
max(max(flow(:,:,2))), min(min(flow(:,:,2)))]; % channel 2
flow_small =... |
github | jjcharles/personalized_pose-master | uncompress_flow_minmax.m | .m | personalized_pose-master/+flow/uncompress_flow_minmax.m | 386 | utf_8 | d0dab4259676d4c79d8ac4c3b277cb59 | %function to read in flow file and uncompress
function flow_large = uncompress_flow_minmax(flow,minmax)
xflow = bsxfun(@times,double(flow(:,:,1,:))/255,permute(minmax(1,1,:)-minmax(1,2,:),[1 2 4 3])) + minmax(1,2);
yflow = bsxfun(@times,double(flow(:,:,2,:))/255,permute(minmax(2,1,:)-minmax(2,2,:),[1 2... |
github | jjcharles/personalized_pose-master | body_part_clustering.m | .m | personalized_pose-master/+cluster/body_part_clustering.m | 4,271 | utf_8 | 6b00ef80fd182df892176a6d3ffb29fa | %function to cluster patches around body parts, based on known joint
%locations
function [part_clusters,detections] = body_part_clustering(opts,videofilename,omodel,detections,patchfilename,cache_folder)
if ~exist('cache_folder','var')
cache_folder = './cache/';
end
if ~exist(cache_folder,'d... |
github | jjcharles/personalized_pose-master | get_part_proposals_with_rot.m | .m | personalized_pose-master/+cluster/get_part_proposals_with_rot.m | 8,928 | utf_8 | 4e22804335f273ed421cc5383b058700 | %function to extract patches from a video given proposed body part
%locations and augments data with rotations
function part_clusters = get_part_proposals_with_rot(opts,videofilename,detections,facepatches,patchwidth)
visualise = false;
if visualise
figure
end
vidobj = Vide... |
github | jjcharles/personalized_pose-master | get_part_proposals.m | .m | personalized_pose-master/+cluster/get_part_proposals.m | 5,280 | utf_8 | adedd8947faa9ac4aec1fa4f76697adf | %function to extract patches from a video given proposed body part
%locations
function part_clusters = get_part_proposals(opts,videofilename,detections,facepatches,patchwidth)
visualise = false;
if visualise
figure
end
vidobj = VideoReader(videofilename);
numparts = si... |
github | jjcharles/personalized_pose-master | label_assignment.m | .m | personalized_pose-master/+cluster/label_assignment.m | 328 | utf_8 | f3ec60e75fba81fbcee3209dc88f3399 | %function to determin cluster assignment given cluster centroids and input
%vectors
function [labels dist] = label_assignment(centroids, y)
index = flann_build_index(centroids,struct('algorithm','kdtree','tree',16));
search_struct = struct('checks',128);
[labels,dist] = flann_search(index,y,1,search_... |
github | jjcharles/personalized_pose-master | exemplarsvm_cluster.m | .m | personalized_pose-master/+cluster/exemplarsvm_cluster.m | 2,023 | utf_8 | 81771229a233be685d3f032401a6be57 | %function to perform exemplar svm clustering
function [labels, dists] = exemplarsvm_cluster(centroids,data,labels)
%thresh is a value between 0 and 1 which will discard data items if they
%are not close to a centroid, higher the thresh, the tighter the control
%data must be scaled between 0 and 1
data = sparse... |
github | jjcharles/personalized_pose-master | load_system_options.m | .m | personalized_pose-master/system/load_system_options.m | 1,556 | utf_8 | bdea367b9a72bcf05b16f1e3aae49da9 | %function to load the correct options files
function [opts,folder] = load_system_options(exp_name,videoname)
switch lower(exp_name)
case 'youtube'
[opts,folder] = load_youtube_options(videoname);
end
function [opts,folder] = load_youtube_options(videoname)
% part_detector_options;
pa... |
github | jjcharles/personalized_pose-master | waitforalljobs.m | .m | personalized_pose-master/system/waitforalljobs.m | 372 | utf_8 | 7cc3afb1a84920145ec5e7c72cf1eb76 | %hangs for all jobs to finish
function waitforalljobs(filename,maxjobs)
joblist = dir(sprintf('%s*',filename.joblist));
fprintf('waiting for %s\n',filename.joblist(1:end-4));
while numel(joblist)<maxjobs
pause(10)
%get list of all running jobs
joblist = dir(sprintf('%s*',fil... |
github | jjcharles/personalized_pose-master | get_background_model.m | .m | personalized_pose-master/system/get_background_model.m | 538 | utf_8 | 8f257aeaebdf8bfbe0bbf5cdd8121f1d | %function to get foreground background segmentation from trained detector
function get_background_model(opts,filename,folder,detections_filename,jobid)
if ~exist(filename.patches,'file')
if jobid == 1
%load a detector
model = oforest.load_forest_from_folder(folder.detector);
%loa... |
github | jjcharles/personalized_pose-master | get_part_locs.m | .m | personalized_pose-master/system/get_part_locs.m | 4,307 | utf_8 | 8c5433477e858b91873835784c40b8bd | %function to return part locations splitting the job over multiple clusters
function detections = get_part_locs(opts,filename,folder_detector,folder_cache,model_scale,jobid,maxjobs)
%model_scale is the amount by which we wish to scale the image size so the
%model runs faster when detecting proposals
if mode... |
github | jjcharles/personalized_pose-master | setjobcompletion.m | .m | personalized_pose-master/system/setjobcompletion.m | 218 | utf_8 | fc9795df848ea3819502881744f1e37c | %function to set job completion tag
function setjobcompletion(joblistfilename,jobid)
jobfilename = sprintf('%s%04d.mat',joblistfilename,jobid);
iscomplete = true;
save(jobfilename,'iscomplete');
|
github | jjcharles/personalized_pose-master | match_body_parts.m | .m | personalized_pose-master/system/match_body_parts.m | 799 | utf_8 | 679203190929da7a06c5939a8153bcd7 | %function to perform clustering
function match_body_parts(opts,filename,folder_detector,cluster_folder,detections,maxjobs)
if ~exist(sprintf('%s%04d.mat',filename.joblist,maxjobs),'file')
%perform matching
if ~exist(filename.unrefined_clusters,'file') && ~exist(filename.clusters,'file')
... |
github | jjcharles/personalized_pose-master | setup_filenames.m | .m | personalized_pose-master/system/setup_filenames.m | 1,015 | utf_8 | cc4482bc822931da2f1b4f48dee4d970 | %new function to setup filenames
function [filename,folder] = setup_filenames(folder, videoname, exp_name)
switch lower(exp_name)
case 'youtube'
filename.video = sprintf('%s%s.mj2',folder.video,videoname);
end
if ~exist(filename.video,'file')
filename.video = sprintf('%s%s.avi',folder.video,videoname)... |
github | jjcharles/personalized_pose-master | waitforremainingjobs.m | .m | personalized_pose-master/system/waitforremainingjobs.m | 376 | utf_8 | 3c0275651892da30dd24f5214ba0f91d | %hangs for all jobs to finish
function waitforremainingjobs(filename,maxjobs)
joblist = dir(sprintf('%s*',filename.joblist));
fprintf('waiting for %s\n',filename.joblist(1:end-4));
while numel(joblist)<(maxjobs-1)
pause(10)
%get list of all running jobs
joblist = dir(sprintf... |
github | jjcharles/personalized_pose-master | get_clustered_manuals.m | .m | personalized_pose-master/system/get_clustered_manuals.m | 3,624 | utf_8 | b1dc6fa1dc11f80c6df56bb619fc23f8 | %function to form clustered manuals
function detections = get_clustered_manuals(opts,filename,detections,feature_type,jobid,maxjobs)
if jobid == 1
joblistfilename = sprintf('%s%04d.mat',filename.joblist,jobid);
if ~exist(joblistfilename,'file')
fprintf('Getting clustered manuals...\n');
... |
github | jjcharles/personalized_pose-master | train_lower_arm_evaluator.m | .m | personalized_pose-master/system/train_lower_arm_evaluator.m | 843 | utf_8 | 3391fbeb6457378aa15b90bc2ed40c61 | %function to perform foreground background detection evaluation on the
%cluster
function train_lower_arm_evaluator(opts,filename,folder,fieldname,evaluator_id,jobid)
%evaluator_id is a string which identifies the evaluation
load(filename.detections);
%runevaluation
if jobid == 1
evaluator_folder = check_di... |
github | jjcharles/personalized_pose-master | get_part_proposals_hpc.m | .m | personalized_pose-master/system/get_part_proposals_hpc.m | 3,353 | utf_8 | 211373f3a2cbbd520a40cac3cccf5213 | %function to get part proposals on the cluster
function part_clusters = get_part_proposals_hpc(opts,filename,detections,jobid,maxjobs,cache_folder)
mainfilename = sprintf('%spartcluster.mat',cache_folder);
if ~exist(mainfilename,'file')
%split up jobs
split = repmat(floor(numel(detections... |
github | jjcharles/personalized_pose-master | setupwaiting.m | .m | personalized_pose-master/system/setupwaiting.m | 289 | utf_8 | d674c7311d2557bf18db01fa96fc3ddd | %setup the file to perform job waiting
function filename = setupwaiting(filename,folder,jobid,maxjobs,tag)
foldername = sprintf('%s%s/',folder.cache,tag);
if ~exist(foldername,'dir'); mkdir(foldername); end
filename.joblist = sprintf('%sjoblist_%s.mat',foldername,tag);
|
github | jjcharles/personalized_pose-master | apply_siftflow.m | .m | personalized_pose-master/system/apply_siftflow.m | 3,671 | utf_8 | 38cd96eb6efd6502b6354cd8329777c8 | %function to perform refinement on the cluster using siftflow
function apply_siftflow(filename,folder,jobid,maxjobs,itr)
%load detections
load(filename.detections);
if ~isfield(detections,'refinement')
if ~exist(sprintf('%s%04d.mat',filename.joblist,jobid),'file')
flow_folder = sprintf('%ssiftflow... |
github | jjcharles/personalized_pose-master | update_initial_detections.m | .m | personalized_pose-master/system/update_initial_detections.m | 3,446 | utf_8 | ec38d5c65018bfed305c70fb1da7ac9d | %function to update the manual detections with new ground truth
function update_initial_detections(filename,olddetections,samplewidth,newfilename,jobid,maxjobs)
if jobid == 1
if ~exist(newfilename,'file')
detections.manual.frameids =[];
detections.manual.locs = [];
detections.manual.... |
github | jjcharles/personalized_pose-master | f_b_evaluation_hpc.m | .m | personalized_pose-master/system/f_b_evaluation_hpc.m | 4,128 | utf_8 | 6935a18d5b8fb559469220fccde9ec2c | %function to perform foreground background detection evaluation on the
%cluster.
%uses svm trained lower arm evaluators where the thresholds were
%learnt on a hold-out validation set - also runs over multiple jobs
function f_b_evaluation_hpc(opts,filename,folder,fieldname,evaluator_id,jobid,maxjobs)
%evaluator_id ... |
github | jjcharles/personalized_pose-master | train_occlusion_detector.m | .m | personalized_pose-master/system/train_occlusion_detector.m | 799 | utf_8 | 35c27b830f18e7a67fb9140bd0be40a9 | %function to train occlusion detector on HPC
function train_occlusion_detector(opts,filename,folder,detections,fieldname,tag,jobid,maxjobs)
folder.occlusiondetector = check_dir(sprintf('%socclusion_detectors',folder.model),true);
filename.occlusiondetector = sprintf('%socclusion_detector_%s.mat',folder.occlusi... |
github | jjcharles/personalized_pose-master | collect_optic_flow.m | .m | personalized_pose-master/system/collect_optic_flow.m | 2,422 | utf_8 | 273fe3ce1febb2952af677adabaabd7c | %function for producing dense optic flow video using DeepFlow
function collect_optic_flow(videoname,exp_name)
%setup filenames and folders
[opts,folder] = load_system_options(exp_name,videoname);
[filename,folder] = setup_filenames(folder, videoname, exp_name);
%create matfile for optical flow output... |
github | jjcharles/personalized_pose-master | get_training_detections_annotation.m | .m | personalized_pose-master/system/get_training_detections_annotation.m | 1,130 | utf_8 | 2797d20377e955ab4822e55cfb080c79 | %function to form clustered manuals
% uses a pose sampler rather than clustering patches
%uses a tag to save the detections
%allows for each tree to sample its own training data
function detections = get_training_detections_annotation(folder,filename,detections,tag,jobid,maxjobs)
temptrain_folder = check_dir(sprin... |
github | jjcharles/personalized_pose-master | apply_occlusion_detection.m | .m | personalized_pose-master/system/apply_occlusion_detection.m | 2,849 | utf_8 | 321f98a1a5daab46f878a5bff24f6623 | %function to apply the occlusion detector on the video (accross HPC nodes)
function apply_occlusion_detection(opts,filename,folder,fieldname,tag,jobid,maxjobs)
%load detections
load(filename.detections);
folder.occlusiondetector = check_dir(sprintf('%socclusion_detectors',folder.model),true);
file... |
github | jjcharles/personalized_pose-master | train_detector_tree.m | .m | personalized_pose-master/system/train_detector_tree.m | 6,034 | utf_8 | 7c965a7bfcbef22480eb25413f39a9b3 | %function to train a part detector
function train_detector_tree(opts, filename, folder, maxtreedepth, jobid, maxjobs, frameids, locs)
savemaxjobs = maxjobs;
if maxjobs == 1
jobidloop = 1:opts.partdetector.model.forest.numtrees;
maxjobs = opts.partdetector.model.forest.numtrees;
else
%return if jobi... |
github | jjcharles/personalized_pose-master | getpatch.m | .m | personalized_pose-master/methods/getpatch.m | 614 | utf_8 | 503c82a1f4dc508bf9e7ccfcb3691dea | %function to get patch from frame
function [patch,bbox] = getpatch(pos,patchwidth,frame)
%if pos out of bounds then bring back to min or max
[M,N,~] = size(frame);
if pos(1)>N; pos(1)=N; end; if pos(1)<1; pos(1)=1; end;
if pos(2)>M; pos(2)=M; end; if pos(2)<1; pos(2)=1; end;
framepadded = ... |
github | jjcharles/personalized_pose-master | plot_skeleton.m | .m | personalized_pose-master/methods/plot_skeleton.m | 3,829 | utf_8 | 8030c3f09e2cf52ae586959e1780c914 | %PLOT_SKELETON - plots skelton of signer on figure
% handle = plot_skeleton(j,opts,handle) j is a 2x7 vector of joints, handle is a struct
% handle.axis
% handle.ula - upper left arm
% handle.ura -
% handle.lla - lower left arm
% handle.lra
% handle.joints(7)
%
% opts.clr = jet(7) = joints
... |
github | jjcharles/personalized_pose-master | showHOG.m | .m | personalized_pose-master/methods/showHOG.m | 1,374 | utf_8 | b004848cdd6833041bde8ba7f946cc71 | % showHOG(w)
%
% Legacy HOG visualization
function out = showHOG(w)
w = w(:, :, 1:9) + w(:, :, 10:18) + w(:, :, 19:27);
w = w / 3;
w = repmat(w, [1 1 3]);
w = padarray(w, [0 0 5], 'post');
% Make pictures of positive and negative weights
bs = 20;
pos = HOGpicture(w, bs);
neg = HOGpicture(-w, bs);
% Put pictures toge... |
github | jjcharles/personalized_pose-master | getbestclusters.m | .m | personalized_pose-master/methods/getbestclusters.m | 1,055 | utf_8 | e494b599e5f80f1bb29f1fd08c4c304d | %function to pick best covering of a set given a support window
function [centroids, clusterids] = getbestclusters(data,windowsize)
max_clusters = min(50,size(data,2));
centroids = [];
clusterids = [];
visualise = false;
for k = 1:max_clusters
isgood = true;
[c,id] = vl_kmeans(double(data),k,'Init... |
github | jjcharles/personalized_pose-master | check_dir.m | .m | personalized_pose-master/methods/check_dir.m | 558 | utf_8 | 5d5ce6a730bf9775cc5d16ec524ea467 | %function to check and correct directory name
function [directory_out,isthere] = check_dir(directory_in,iscreate)
if ~exist('iscreate','var')
iscreate = false;
end
directory_out = directory_in;
directory_out(directory_out=='\') = '/'; %change to unix style
if directory_out(... |
github | jjcharles/personalized_pose-master | siftflow.m | .m | personalized_pose-master/methods/siftflow.m | 1,026 | utf_8 | cffa6f461d48b84ec8e48fc3d94caaf7 | %output flow field based on SIFT flow and energy
function [flow, energy] = siftflow(im1,im2,mask)
%mask is actually 2d x, y coordinates within the image space and not a
%binary image
%PARAMETERS
cellsize=3;
gridspacing=1;
flow = zeros(size(im1,1),size(im1,2),2); %sift flow turned off!!!
... |
github | jjcharles/personalized_pose-master | show_locs.m | .m | personalized_pose-master/methods/show_locs.m | 1,540 | utf_8 | 38f256f330c8cc521e7aa9d0a36a10bc | %function to visualise video and joint locations
function show_locs(videofilename,imscale,frameids,locs,waittime)
if ~exist('waittime','var'); waittime = 0; end
videofolder = './videos/annotation/';
if ~exist(videofolder,'dir'); mkdir(videofolder); end
vidobj = VideoRead... |
github | jjcharles/personalized_pose-master | sample_uniformally.m | .m | personalized_pose-master/methods/sample_uniformally.m | 2,263 | utf_8 | 0b575a9c3638e16d62c2e9cde55776ef | %function to sample uniformally from a set of points from R^n using a
%maximal coverage algorithm with window width as a parameter. I.e. it
%covers the input set with the minimum number of cirular windows with
%diameter given by 'window_width'. It then samples uniformally with
%replacement from each covered region.... |
github | jjcharles/personalized_pose-master | train_svm_classifier.m | .m | personalized_pose-master/methods/train_svm_classifier.m | 4,023 | utf_8 | bc45b34c564dce63f8cc7af21ee99708 | %function to train and tune an svm classifier
function [svmmodel, best_thresh] = train_svm_classifier(labels,data,required_sensitivity)
labels = labels(:);
%first get c value right
trialc = 100./10.^[1:15];
c_score = zeros(1,numel(trialc));
for cc = 1:numel(trialc)
%split data into ... |
github | jjcharles/personalized_pose-master | show_skeleton.m | .m | personalized_pose-master/methods/show_skeleton.m | 823 | utf_8 | ef67b1c242ee5c3a6809ffe4aff76666 | %function to visualise video and joint locations
function show_skeleton(videofilename,imscale,frameids,locs,secs)
vidobj = VideoReader(videofilename);
figure
img = imresize(read(vidobj,frameids(1)),imscale);
h_img = imagesc(img); axis image; hold on
h_plot = plot_skeleton(zeros(2,size(locs,2... |
github | jjcharles/personalized_pose-master | flann_search.m | .m | personalized_pose-master/methods/flann/flann_search.m | 3,564 | utf_8 | a5a9b7cb6bc8b49d8f0f6c2737c17608 | %Copyright 2008-2009 Marius Muja (mariusm@cs.ubc.ca). All rights reserved.
%Copyright 2008-2009 David G. Lowe (lowe@cs.ubc.ca). All rights reserved.
%
%THE BSD LICENSE
%
%Redistribution and use in source and binary forms, with or without
%modification, are permitted provided that the following conditions
%are met:
%
... |
github | jjcharles/personalized_pose-master | draw.m | .m | personalized_pose-master/+pfilter/draw.m | 1,130 | utf_8 | 4aa8a3f143799559e3ce215a3605e6c9 | %function to plot particles on img input
function hndl = draw(particle,img,hndl)
if nargin < 3
hndl{1} = gcf;
hndl{2} = imagesc(img);
axis image
hold on
else
figure(hndl{1});
set(hndl{2},'cdata',img);
end
X = cat(1,partic... |
github | jjcharles/personalized_pose-master | get_prediction.m | .m | personalized_pose-master/+pfilter/get_prediction.m | 1,024 | utf_8 | 739fe0f7a7048ba4956f6fe196a8a24a | %returns predictions from particles
function [P, conf] = get_prediction(particle,img_size)
x = cat(1,particle(:).x);
y = cat(1,particle(:).y);
x = round(x);
y = round(y);
bbox_in = [min(x),min(y),max(x)-min(x)+1,max(y)-min(y)+1];
x = x-bbox_in(1)+1;
y = y-bbox_in(2)+1... |
github | jjcharles/personalized_pose-master | resample.m | .m | personalized_pose-master/+pfilter/resample.m | 850 | utf_8 | 51e1bf972ee99750861de86b7724856d | %function to resample particles
function particle = resample(particle,N,conf)
%N is number to resample (optional) default is same number as number of
%input particles
%always need at least number of input particles so we resample more from
%conf
if nargin < 2
N = numel(particle);
else
... |
github | jjcharles/personalized_pose-master | quant_segcp.m | .m | personalized_pose-master/+features/quant_segcp.m | 1,017 | utf_8 | 8416e1917ddd05a41f07a1615ead964b | function feat = quant_segcp(opts,rgbimg,patches,seg)
%load lookup table for quantisation
[l_info.bin2clr, l_info.clr2bin] = colour2bin();
segcp_feat = features.segcp(opts,rgbimg,patches,seg);
%quantise frame IMPORTANT
bin_idx = l_info.clr2bin(segcp_feat+1);
feat = uint8(l_info.bin2... |
github | jjcharles/personalized_pose-master | rgb_attenuated.m | .m | personalized_pose-master/+features/rgb_attenuated.m | 349 | utf_8 | 6363e860f5ff6a1ff51b71a0cd58991a | %attenuated foreground
function feat = rgb_attenuated(opts,rgbimg,patches)
cpimg = features.cp(opts,rgbimg,patches);
%get forground probability
fp = sum(double(cpimg(:,:,1:2)),3)./sum(double(cpimg),3);
%weight the rgbimg according to foreground
feat = uint8(bsxfun(@times,... |
github | jjcharles/personalized_pose-master | cp.m | .m | personalized_pose-master/+features/cp.m | 2,703 | utf_8 | bb423da513e10f699736d5224e443e83 | %computer segcp feature from input rgb and face and torso patches and
%segmentation
function [feat,colour_hist] = cp(opts,rgbimg,patches)
if isfield(patches{1},'colourhist')
colour_hist = patches{1}.colourhist;
else
ref.face = [];
ref.torso = [];
ref.back = [];
... |
github | jjcharles/personalized_pose-master | weighted_hog.m | .m | personalized_pose-master/+features/weighted_hog.m | 726 | utf_8 | 517fa46d2c93bee66588e7c36400fde0 | %function to create a posterior weighted HOG feature
function [weighted_hog,minvalue,maxvalue] = weighted_hog(rgbimg,cpimg,cellsize)
%get hog feature
hog = features.hog(rgbimg,cellsize);
hog = double(hog);
%get forground probability
cpd = cpimg; for i = 1:3; cpd(:,:,i) = medfilt2(cpimg(:,:... |
github | jjcharles/personalized_pose-master | hog.m | .m | personalized_pose-master/+features/hog.m | 116 | utf_8 | c629d98949074883c6d694d19c6d105c | %function to produce hog features
function feat = hog(img,cellsize)
feat = hogfeat(im2double(img),cellsize); |
github | jjcharles/personalized_pose-master | segcp.m | .m | personalized_pose-master/+features/segcp.m | 1,797 | utf_8 | 8880caeb6462b72c3dc896aa162a8ef8 | %computer segcp feature from input rgb and face and torso patches and
%segmentation
function feat = segcp(opts,rgbimg,patches,seg)
seg = double(seg);
ref.face = [];
ref.torso = [];
for template_id = 1:numel(patches)
if (~isempty(patches{template_id}.face))
face = patche... |
github | jjcharles/personalized_pose-master | hardsegcp.m | .m | personalized_pose-master/+features/hardsegcp.m | 463 | utf_8 | ab34895cf24985322d818820509295a8 | %computes hardsegcp feature from rgb input frame
function feat = hardsegcp(opts, rgbimg, patches, seg, bbox)
img = features.segcp(opts,rgbimg,patches,seg);
disk = strel('disk',2);
[~,channel_id] = max(img,[],3);
seg_channel = imopen(channel_id==1,disk);
dist = bwdist(seg_channel);
... |
github | jjcharles/personalized_pose-master | img2can.m | .m | personalized_pose-master/+shape/img2can.m | 1,127 | utf_8 | 28231d521fc7edc57fbbd0f350c566f1 | %function to extract rectangle from image provided with
%-image
%-anchor points on rectangle as percentage of rectangle height
%-2x2d coordinates in the image
%-input width of rectangle
%-output width and height
%Output is a canonicalised patch from the input image when converted to
%colour_space either 'RGB' or... |
github | spkrafft/trex-master | TREX.m | .m | trex-master/TREX.m | 4,845 | utf_8 | 00ce0259cc8d8f76d9414b04ead27abf | function varargout = TREX(varargin)
% TREX MATLAB code for TREX.fig
% TREX, by itself, creates a new TREX or raises the existing
% singleton*.
%
% H = TREX returns the handle to a new TREX or the handle to
% the existing singleton*.
%
% TREX('CALLBACK',hObject,eventData,h,...) calls the local
%... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.