edgesDetectDL.m

function [E,O,inds,segs] = edgesDetectDL( I, model )
% Detect edges in image by deep contour features.
% The code is modified from Structured Edge Detection Toolbox



% get parameters
opts=model.opts; opts.nTreesEval=min(opts.nTreesEval,opts.nTrees);
if(~isfield(opts,'sharpen')), opts.sharpen=0; end
if(~isfield(model,'segs')), model.segs=[]; model.nSegs=[]; end
opts.stride=max(opts.stride,opts.shrink); model.opts=opts;
testFeatFir = [opts.modelDir '/feat/'];

if( opts.multiscale )
  % if multiscale run edgesDetect multiple times
  ss=2.^(-1:1); k=length(ss); inds=cell(1,k); segs=inds;
  siz=size(I); model.opts.multiscale=0; model.opts.nms=0; E=0;
  for i=1:k, s=ss(i); 
      I1=imResample(I,s);
      if(s~=1)
          model.oriIm = 0;
      else
          model.oriIm = 1;
      end
    if(nargout<4), 
        [E1,inds{i}]=edgesDetectDL(I1,model);
    else
        [E1,inds{i},segs{i}]=edgesDetectDL(I1,model); 
    end
    E=E+imResample(E1,siz(1:2));
  end; E=E/k; model.opts=opts;
  
else
  % pad image, making divisible by 4
  siz=size(I); r=opts.imWidth/2; p=[r r r r];
  p([2 4])=p([2 4])+mod(4-mod(siz(1:2)+2*r,4),4);
  I = imPad(I,p,'symmetric');
  % compute features and apply forest to image
  if(isfield(model,'oriIm') && model.oriIm ~= 1)
    deepFeat = edgesChnsDL2(I, opts);
  else
      if exist([testFeatFir model.imgId '_deepFeat.mat'], 'file');
        load([testFeatFir model.imgId '_deepFeat.mat']);
      else
        deepFeat = edgesChnsDL2(I, opts);
        save([testFeatFir model.imgId '_deepFeat.mat'], 'deepFeat');
      end
  end
  if(isfield(opts, 'selected_dims') && ~isempty(opts.selected_dims))
    deepFeat = deepFeat(:, :, opts.selected_dims);
  end
%   [deepFeat,~] = edgesChns( I, opts );
%   deepFeat = deepFeat(:, :, 4);
  s=opts.sharpen; if(s), I=convTri(rgbConvert(I,'rgb'),1); end
  if(nargout<4), [E,inds] = edgesDetectMexCNNFeats(model,I,deepFeat);
  else [E,inds,segs] = edgesDetectMexCNNFeats(model,I,deepFeat); end
  
  % normalize and finalize edge maps
  t=opts.stride^2/opts.gtWidth^2/opts.nTreesEval; r=opts.gtWidth/2;
  if(s==0), t=t*2; elseif(s==1), t=t*1.8; else t=t*1.66; end
  E=E(1+r:siz(1)+r,1+r:siz(2)+r,:)*t; E=convTri(E,1);
end

% compute approximate orientation O from edges E
if( opts.nms==-1 ), O=[]; elseif( nargout>1 || opts.nms )
  [Ox,Oy]=gradient2(convTri(E,4));
  [Oxx,~]=gradient2(Ox); [Oxy,Oyy]=gradient2(Oy);
  O=mod(atan(Oyy.*sign(-Oxy)./(Oxx+1e-5)),pi);
end

% perform nms
if( opts.nms>0 ), E=edgesNmsMex(E,O,1,5,1.01,opts.nThreads); end

Border=5;        
for s=1:Border,
    E([s end-s+1],:,:)=E([s end-s+1],:,:)*(s-1)/Border;
end
for s=1:Border,
    E(:,[s end-s+1],:)=E(:,[s end-s+1],:)*(s-1)/Border;
end

end