init.m

clear;
format compact; format long g;

global config;
config.CLASSES={'aeroplane','bicycle','bird','boat','bottle','bus','car','cat','chair','cow','diningtable','dog','horse','motorbike','person','pottedplant','sheep','sofa','train','tvmonitor'};

config.DEBUG=0; % debug and verbosity level

config.POSELET_SELECTION_SMOOTH_T = 5;

% Parameters that define the configuration distance between two patches and
% the poselet example selection algorithm
config.MIN_ROT_THRESH = pi*3/4;        % a sample that is rotated +/- more than this is discarded
config.MAX_ZOOM_THRESH = 4;            % maximum zoom for a sample to be kept (4 means each source pixel can be magnified at most x4)
config.VISUAL_DIST_WEIGHT = 0.1;       % Weight of visual distance relative to Procruster's distance
config.USE_PHOG = false;
config.USE_SEGMENT_DISTANCE = false;
config.ERR_THRESH = 2;                 % what fraction of the samples to keep per each part (based on configuration proximity)

% HOG parameters (set according to the paper N.Dalal and B.Triggs, "Histograms of Oriented Gradients
% for Human Detection" CVPR 2005)
config.HOG_CELL_DIMS = [16 16 180];
config.NUM_HOG_BINS = [2 2 9];
config.SKIN_CELL_DIMS = [2 2];
config.HOG_WTSCALE = 2;
config.HOG_NORM_EPS = 1;
config.HOG_NORM_EPS2 = 0.01;
config.HOG_NORM_MAXVAL = 0.2;
config.HOG_NO_GAUSSIAN_WEIGHT=false;
config.USE_PHOG=false;

% Scanning parameters
config.PYRAMID_SCALE_RATIO = 1.1;
config.DETECTION_IMG_MIN_NUM_PIX = 1000^2;  % if the number of pixels in a detection image is < DETECTION_IMG_SIDE^2, scales up the image to meet that threshold
config.DETECTION_IMG_MAX_NUM_PIX = 1500^2;  
config.DETECT_SVM_THRESH = 0;               % higher = more more precision, less recall
config.MAX_AGGLOMERATIVE_CLUSTER_ELEMS = 500;
config.DETECT_MAX_HITS_PER_SCALE_PER_POSELET = inf; 

% Poselet clustering parameters
config.HYP_CLUSTER_THRESH = 5; %400;    % KL-distance between poselet hits to be considered in the same cluster. Used for personalized clustering of big Qs
config.GREEDY_CLUSTER_THRESH = 5;
% if ~isfield(config,'GREEDY_CLUSTER_THRESH')
%     config.GREEDY_CLUSTER_THRESH = 1;   % KL-distance between poselet hits to be considered in the same cluster. Used in greedy clustering
% end
config.HYP_CLUSTER_MAXNUM = 100;        % Max number of clusters in an image
config.CLUSTER_HITS_CUTOFF=0.6;         % clustering threshold for bounds hypotheses
config.HYPOTHESIS_PRIOR_VAR = 1;                % value of prior on the variance of keypoint distribution
config.HYPOTHESIS_PRIOR_VARIANCE_WEIGHT = 1;    % weight of prior on the variance of keypoint distribution
config.KL_USE_WEIGHTED_DISTANCE = false;        % If using KL-divergence, do we give a separate weight for each keypoint
config.CLUSTER_BOUNDS_DIST_TYPE=0;              % type of distance metric for clustering poselets.


config.USE_MEX_HOG=false;                % disable this to use Matlab version instead of mex file for HOG
config.USE_MEX_RESIZE=false;             % disable this to use Matlab version instead of mex file for imresize

% Other parameters
config.TORSO_ASPECT_RATIO = 1.5;        % height/width of torsos
config.CROP_PREDICTED_OBJ_BOUNDS_TO_IMG=true;


for i=1:length(config.CLASSES)
    config_file = sprintf('config_%s',config.CLASSES{i});
    if exist(config_file,'file')
       config.K(i) = eval(config_file); 
       disp(sprintf('configuring %s',config.CLASSES{i}));
    end        
end

clear i config_file;