computeSTpairCost.m

function cost_edges = computeSTpairCost(detections)
addpath(genpath('/BS/pedestrian-detection-tracking/work/project/3rd/liblinear-1.94/'));

%% pairwise logit model
load /BS/joint-multicut/work/MOT-data/models/pairwise/trainSeq_ADL-Rundle-6/logit_pairwise_position-scale_linear-quadratic_posIou-0.5_negExType-mixed-negIou-0.3_limit_1_L2_frameOffset_1.mat;

num_frame = length(detections);
cost_edges = cell(num_frame);
for i = 1:num_frame-1
    fprintf('Compute pairwise cost for frame:  %d ', i)
    det_1 = detections{i};
    
    % computing pairwise cost between frame i and frame j
    j = i+1;
    
    fprintf(' and  %d', j)
    det_2 = detections{j};
    pairwise_prob = compute_pairwise_cost_block(det_1,det_2,model,normalize_ratio);
    pairwise_cost = log((1-pairwise_prob)./pairwise_prob);
    cost_edges{i,j} = pairwise_cost(:);
end
end

function pairwise_prob = compute_pairwise_cost_block(b1,b2,model,normalize_ratio)

num_det_1 = size(b1,1);
num_det_2 = size(b2,1);

pairwise_prob = zeros(num_det_2,num_det_1);

for d1 = 1 : num_det_1
    xc_1 = (b1(d1,1)+b1(d1,3))/2;
    yc_1 = (b1(d1,2)+b1(d1,4))/2;
    h_1 = b1(d1,4)-b1(d1,2);
    w_1 = b1(d1,3)-b1(d1,1);
    
    idx =  1:num_det_2;
    xc_list = (b2(idx,1) + b2(idx,3))/2;
    yc_list = (b2(idx,2) + b2(idx,4))/2;
    h_list = b2(idx,4) - b2(idx,2);
    w_list = b2(idx,3) - b2(idx,1);
    
    [~, ~, iou_n] = iou(xc_1, yc_1, h_1, w_1, xc_list, yc_list, h_list, w_list);
    
    f = [];
    
    ave_h = ((b2(idx,4) - b2(idx,2)) + h_1)/2;
    offset_h = (h_list-h_1)./ave_h; % h' -h
    x = (xc_list - xc_1)./ave_h; %x' - x
    y = (yc_list - yc_1)./ave_h; %y' - y
    
    
    f(:,1) =  abs(x);
    f(:,2) =  abs(y);
    f(:,3) =  iou_n;
    f(:,4) =  abs(offset_h);
    
    %% for quadratic kernel
    
    [exp_num, dim] = size(f);
    
    f_quad = zeros(exp_num, (dim-1)*dim/2);
    dim_count = 1;
    for m = 2:dim
        for n = 2:dim
            f_quad(:,dim_count) = f(:,m).*f(:,n);
            dim_count = dim_count + 1;
        end
    end
    f = [f f_quad];
    
    %%
    f =  normalize_feature(f, normalize_ratio);
    
    f = sparse(double(f));
    label = ones(size(f,1),1);
    [~, ~,pro_estimate] = predict(label, f, model,  '-b 1 -q');
    
    pairwise_prob(:,d1) = pro_estimate(:,1);
end
end
%