Struct contents reference from a non-struct array object.

This is the code that I use ,but I get this erorr. @Please, can you help me.

f

unction model = stTrain( varargin )
% See stDemo for a full demo that include both traininga and application.
%
% Pre-trained models can be downloaded from:
% http://people.csail.mit.edu/lim/lzd_cvpr2013/st_data.tgz
%
% Please cite the following paper if you end up using the code:

% USAGE
%  model = stTrain( opts )
%
% INPUTS
%  opts       - parameters (struct or name/value pairs)
%   (1) parameters for model and data:
%   .nClusters  - [150] number of clusters to train with
%   .nTrees     - [25] number of trees in forest to train
%   .radius     - [17] radius of sketch token patches
%   .nPos       - [1000] number of positive patches per cluster
%   .nNeg       - [800] number of negative patches per image
%   .negDist    - [2] distance from closest contour defining a negative
%   .minCount   - [4] minimum number of training examples per node
%   (2) parameters for features:
%   .nCells     - [5] number of self similarity cells
%   .normRad    - [5] normalization radius (see gradientMag)
%   .normConst  - [.01] normalization constant (see gradientMag)
%   .nOrients   - [4 4 0] number of orientations for each channel set
%   .sigmas     - [0 1.5 5] gaussian blur for each channel set
%   .chnsSmooth - [2] radius for channel smoothing (using convTri)
%   .fracFtrs   - [1] fraction of features to use to train each tree
%   (3) other parameters:
%   .seed       - [1] seed for random stream (for reproducibility)
%   .modelDir   - ['models/'] target directory for storing models
%   .modelFnm   - ['model'] model filename
%   .clusterFnm - ['clusters.mat'] file containing cluster info
%   .bsdsDir    - ['BSR/BSDS500/data/'] location of BSDS dataset
%
% OUTPUTS
%  model      - trained sketch token detector w the following fields
%   .trees      - learned forest model struct array (see forestTrain)
%   .opts       - input parameters and constants
%   .clusters   - actual cluster centers used to learn tokens
%
    % get default parameters
   
  
    dfs={'nClusters',150, 'nTrees',25, 'radius',17, 'nPos',1000, 'nNeg',800,...
        'negDist',2, 'minCount',4, 'nCells',5, 'normRad',5, 'normConst',0.01, ...
        'nOrients',[4 4 0], 'sigmas',[0 1.5 5], 'chnsSmooth',2, 'fracFtrs',1, ...
        'seed',1, 'modelDir','models/', 'modelFnm','model', ...
        'clusterFnm','clusters.mat', 'bsdsDir','BSR/BSDS500/data/'};
    opts = getPrmDflt(varargin,dfs,1);

    % if forest exists load it and return
    cd(fileparts(mfilename('fullpath')));
    forestDir = [opts.modelDir '/forest/'];
    forestFn = [forestDir opts.modelFnm];
    if exist([forestFn '.mat'], 'file')
        load([forestFn '.mat']);
        return;
    end

    % compute constants and store in opts
    nTrees=opts.nTrees;
    nCells=opts.nCells;
    
    patchSiz=opts.radius*2+1;
    opts.patchSiz=patchSiz;
    
    nChns = size(stChns(ones(2,2,3),opts),3);
    opts.nChns=nChns;
    
    opts.nChnFtrs = patchSiz*patchSiz*nChns;
    opts.nSimFtrs = (nCells*nCells)*(nCells*nCells-1)/2*nChns;
    opts.nTotFtrs = opts.nChnFtrs + opts.nSimFtrs;
    opts.cellRad = round(patchSiz/nCells/2);
    tmp=opts.cellRad*2+1;
    opts.cellStep = tmp-ceil((nCells*tmp-patchSiz)/(nCells-1)); disp(opts);
    assert( (nCells == 0) || (mod(nCells,2)==1 && (nCells-1)*opts.cellStep+tmp <= patchSiz ));

    % generate stream for reproducibility of model
    stream=RandStream('mrg32k3a','Seed',opts.seed);

    % train nTrees random trees (can be trained with parfor if enough memory)
    for i=1:nTrees
        stTrainTree( opts, stream, i );
    end

    % accumulate trees and merge into final model
    treeFn = [opts.modelDir '/tree/' opts.modelFnm '_tree'];
    for i=1:nTrees
        t=load([treeFn int2str2(i,3) '.mat'],'tree');
        t=t.tree;
        if (i==1)
            trees=t(ones(1,nTrees));
        else
            trees(i)=t;
        end
    end
    nNodes=0;
    for i=1:nTrees
        nNodes=max(nNodes,size(trees(i).fids,1));
    end
    model.thrs=zeros(nNodes,nTrees,'single');
    Z=zeros(nNodes,nTrees,'uint32');
    model.fids=Z;
    model.child=Z;
    model.count=Z;
    model.depth=Z;
    model.distr=zeros(nNodes,size(trees(1).distr,2),nTrees,'single');
    for i=1:nTrees, tree=trees(i); nNodes1=size(tree.fids,1);
        model.fids(1:nNodes1,i) = tree.fids;
        model.thrs(1:nNodes1,i) = tree.thrs;
        model.child(1:nNodes1,i) = tree.child;
        model.distr(1:nNodes1,:,i) = tree.distr;
        model.count(1:nNodes1,i) = tree.count;
        model.depth(1:nNodes1,i) = tree.depth;
    end
    model.distr = permute(model.distr, [2 1 3]);
    
    clusters=load(opts.clusterFnm);
    clusters=clusters.clusters;
    
    model.opts = opts;
    model.clusters=clusters.clusters;
    if ~exist(forestDir,'dir')
        mkdir(forestDir);
    end
    save([forestFn '.mat'], 'model', '-v7.3');

end

function stTrainTree( opts, stream, treeInd )
% Train a single tree in forest model.

    % location of ground truth
    trnImgDir = [opts.bsdsDir '/images/train/'];
    trnGtDir = [opts.bsdsDir '/groundTruth/train/'];
    imgIds=dir([trnImgDir '*.png']);
    imgIds={imgIds.name};
    nImgs=length(imgIds);
    for i=1:nImgs,
        imgIds{i}=imgIds{i}(1:end-4);
    end

    % extract commonly used options
    radius=opts.radius;
    patchSiz=opts.patchSiz;
    nChns=opts.nChns;
    nTotFtrs=opts.nTotFtrs;
    nClusters=opts.nClusters;
    nPos=opts.nPos;
    nNeg=opts.nNeg;

    % finalize setup
    treeDir = [opts.modelDir '/tree/'];
    treeFn = [treeDir opts.modelFnm '_tree'];
    if exist([treeFn int2str2(treeInd,3) '.mat'],'file')
        return;
    end
    fprintf('\n-------------------------------------------\n');
    fprintf('Training tree %d of %d\n',treeInd,opts.nTrees);
    tStart=clock;

    % set global stream to stream with given substream (will undo at end)
    streamOrig = RandStream.getGlobalStream();
    set(stream,'Substream',treeInd);
    RandStream.setGlobalStream( stream );

    % sample nPos positive patch locations per cluster
    clstr=load(opts.clusterFnm);
    clstr=clstr.clusters;
    for i = 1:nClusters
        if i==1
            centers=[];
        end
        ids = find(clstr.clusterId == i);
        ids = ids(randperm(length(ids),min(nPos,length(ids))));
        centers = [centers; [clstr.x(ids),clstr.y(ids),clstr.imId(ids),...
            clstr.clusterId(ids),clstr.gtId(ids)]]; %#ok<AGROW>
    end

    % collect positive and negative patches and compute features
    fids=sort(randperm(nTotFtrs,round(nTotFtrs*opts.fracFtrs)));
    k = size(centers,1)+nNeg*nImgs;
    ftrs = zeros(k,length(fids),'single');
    labels = zeros(k,1); k = 0;
    tid = ticStatus('Collecting data',1,1);
    for i = 1:nImgs
        % get image and compute channels
        gt=load([trnGtDir imgIds{i} '.mat']);
        gt=gt.groundTruth;
        
        I = imread([trnImgDir imgIds{i} '.png']);
        I = imPad(I,radius,'symmetric');
        chns = stChns(I,opts);
        
        % sample positive patch locations
        centers1=centers(centers(:,3)==i,:);
        lbls1=centers1(:,4);
        xy1=single(centers1(:,[1 2]));
        %///////////////////////////////////////////////////////////////
        % get image and compute channels
       
        % sample negative patch locations
        M=false(size(I,1)-2*radius,size(I,2)-2*radius);
        nGt=length(gt);
        for j=1:nGt 
            M1=gt{j}.Boundaries;
            % M1=handle.gt{j}.String;
            if ~isempty(M1)
                M=M | M1;
            end
        end
        M(bwdist(M)<opts.negDist)=1;
        M=~M;
        M([1:radius end-radius:end],:)=0;
        M(:,[1:radius end-radius:end])=0;
        [y,x]=find(M);
        k1=min(length(y),nNeg);
        
        rp=randperm(length(y),k1);
        y=y(rp);
        x=x(rp);
        xy0=[x y];
        lbls0=ones(k1,1)*(nClusters+1);
        
        % crop patches
        xy=[xy1; xy0];
        lbls=[lbls1; lbls0];
        k1=length(lbls);
        ps=zeros(patchSiz,patchSiz,nChns,k1,'single');
        p=patchSiz-1;
        for j=1:k1
            ps(:,:,:,j)=chns(xy(j,2):xy(j,2)+p,xy(j,1):xy(j,1)+p,:);
        end
        if(0), montage2(squeeze(ps(:,:,1,:))); drawnow; end
        
        % compute features and store
        ftrs1=[reshape(ps,[],k1)' stComputeSimFtrs(ps,opts)];
        ftrs(k+1:k+k1,:) = ftrs1(:,fids);
        labels(k+1:k+k1) = lbls;
        k=k+k1;
        
        tocStatus(tid,i/nImgs);
    end
    if k<size(ftrs,1)
        ftrs=ftrs(1:k,:);
        labels=labels(1:k);
    end

    % train sketch token classifier (random decision tree)
    tree=forestTrain(ftrs,labels,'maxDepth',999);
    tree.fids(tree.child>0) = fids(tree.fids(tree.child>0)+1)-1;
    tree=pruneTree(tree,opts.minCount); %#ok<NASGU>
    if ~exist(treeDir,'dir')
        mkdir(treeDir);
    end
    save([treeFn int2str2(treeInd,3) '.mat'],'tree');
    e=etime(clock,tStart);
    fprintf('Training of tree %d complete (time=%.1fs).\n',treeInd,e);
    RandStream.setGlobalStream( streamOrig );

end

function tree = pruneTree( tree, minCount )
% Prune all nodes whose count is less than minCount.

    % mark all internal nodes if either child has count<=minCount
    mark = [0; tree.count<=minCount];
    mark = mark(tree.child+1) | mark(tree.child+2);
    
    % list of nodes to be discarded / kept
    disc=tree.child(mark);
    disc=[disc; disc+1];
    n=length(tree.fids);
    keep=1:n;
    keep(disc)=[];
    
    % prune tree
    tree.fids=tree.fids(keep);
    tree.thrs=tree.thrs(keep);
    tree.child=tree.child(keep);
    tree.distr=tree.distr(keep,:);
    tree.count=tree.count(keep);
    tree.depth=tree.depth(keep);
    assert(all(tree.count>minCount))
    
    % re-index children
    route=zeros(1,n);
    route(keep)=1:length(keep);
    tree.child(tree.child>0) = route(tree.child(tree.child>0));
end

function ftrs = stComputeSimFtrs( chns, opts )
% Compute self-similarity features.
    n=opts.nCells;
    if(n==0),
        ftrs=[];
        return;
    end
    nSimFtrs=opts.nSimFtrs;
    nChns=opts.nChns;
    m=size(chns,4);
    
    inds = ((1:n)-(n+1)/2)*opts.cellStep+opts.radius+1;
    chns=reshape(chns,opts.patchSiz,opts.patchSiz,nChns*m);
    chns=convBox(chns,opts.cellRad);
    chns=reshape(chns(inds,inds,:,:),n*n,nChns,m);
    ftrs=zeros(nSimFtrs/nChns,nChns,m,'single');
    k=0;
    for i=1:n*n-1
        k1=n*n-i;
        ftrs(k+1:k+k1,:,:)=chns(1:end-i,:,:)-chns(i+1:end,:,:);
        k=k+k1;
    end
    ftrs = reshape(ftrs,nSimFtrs,m)';
    % % For m=1, the above should be identical to the following:
    % [cids1,cids2]=computeCids(size(chns),opts); % see stDetect.m
    % chns=convBox(chns,opts.cellRad); k=opts.nChnFtrs;
    % cids1=cids1(k+1:end)-k+1; cids2=cids2(k+1:end)-k+1;
    % ftrs=chns(cids1)-chns(cids2);
end

This is the erorr

Struct contents reference from a non-struct array object.

Error in stTrain>stTrainTree (line 225)
            M1=gt{j}.Boundaries;

Error in stTrain (line 102)
        stTrainTree( opts, stream, i );

[amal alhait]

This is the erorr

Struct contents reference from a non-struct array object.
Error in stTrain>stTrainTree (line 225)
            M1=gt{j}.Boundaries;
Error in stTrain (line 102)
        stTrainTree( opts, stream, i );

[Stephen23]

The relevant lines of code are:

        gt=load([trnGtDir imgIds{i} '.mat']);
        gt=gt.groundTruth;
   ...
            M1=gt{j}.Boundaries;

Apparently the loaded .mat file contains a cell array named groundTruth, but this cell array does not contain a structure.

[amal alhait]

I really appreciate your reply

Actually I've created this groundTruth from the following code

s.groundTruth={'D:\Master Degree\SketchTokens\BSR\BSDS500\data\images\train\1.png',[100 73]};

save('1.mat','-struct','s');

So these steps I repeated them to create .mat files for all the images. After creating .mat files it comes with that error.

[Stephen23]

"Actually I've created this groundTruth from the following code"

s.groundTruth={'D:\Master Degree\SketchTokens\BSR\BSDS500\data\images\train\1.png',[100 73]};
save('1.mat','-struct','s');

Those steps willl create a .mat file containing a 2x1 cell array named groundTruth, but none of its cells contain any structures (instead its cells contain a char vector and a numeric vector).

[amal alhait]

Exactly, but how I can solve this issue. I really got tired by trying and nothing works with me

[amal alhait]

Answer 1

[Stephen23]

The code clearly expects the cell array groundTruth to contain a structure, which must have the fields Boundaries and String (and possibly others). Does your cell array contain any structures? No, it does not (it contains a numeric vector and a character vector).

So create your cell array to contain the structures that that code requires:

groundTruth{1} = struct('Boundaries',[0,1],'String','hello.png');
groundTruth{2} = struct('Boundaries',[3,4],'String','world.png');
...
save(..., 'groundTruth')

Better designed code would have used a non-scalar structure:

https://www.mathworks.com/help/matlab/matlab_prog/access-multiple-elements-of-a-nonscalar-struct-array.html

[amal alhait]

Thank you sooooo much this problem solved, but other problem appears

[amal alhait]

Thank you sooooo much this problem solved, but other problem appears

[amal alhait]

I have done this

>> save('1.mat','groundTruth')
>>groundTruth{2}=struct('Boundaries',[3,4],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\2.png' );
>> save('2.mat','groundTruth')
>>groundTruth{3}=struct('Boundaries',[3,4],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\3.png' );
>> save('3.mat','groundTruth')
>>groundTruth{4}=struct('Boundaries',[7,8],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\4.png' );
>> save('4.mat','groundTruth')
>>groundTruth{5}=struct('Boundaries',[9,10],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\5.png' );
>> save('5.mat','groundTruth')
>>groundTruth{6}=struct('Boundaries',[11,12],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\6.png' );
>> save('6.mat','groundTruth')
>>groundTruth{7}=struct('Boundaries',[13,14],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\7.png' );
>> save('7.mat','groundTruth')
>>groundTruth{8}=struct('Boundaries',[15,16],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\8.png' );
>> save('8.mat','groundTruth')
>>groundTruth{9}=struct('Boundaries',[17,18],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\9.png' );
>> save('9.mat','groundTruth')
>>groundTruth{10}=struct('Boundaries',[19,20],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\10.png' );
>> save('10.mat','groundTruth')
>>groundTruth{11}=struct('Boundaries',[21,22],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\11.png' );
>> save('11.mat','groundTruth')
>>groundTruth{12}=struct('Boundaries',[23,24],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\12.png' );
>> save('12.mat','groundTruth')
>>groundTruth{13}=struct('Boundaries',[25,26],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\13.png' );
>> save('13.mat','groundTruth')
>>groundTruth{14}=struct('Boundaries',[27,28],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\14.png' );
>> save('14.mat','groundTruth')
>>groundTruth{15}=struct('Boundaries',[29,30],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\15.png' );
>> save('15.mat','groundTruth')
>>groundTruth{16}=struct('Boundaries',[30,31],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\16.png' );
>> save('16.mat','groundTruth')
>>groundTruth{17}=struct('Boundaries',[31,32],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\17.png' );
>> save('17.mat','groundTruth')
>>groundTruth{18}=struct('Boundaries',[33,34],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\18.png' );
>> save('18.mat','groundTruth')
>>groundTruth{19}=struct('Boundaries',[35,36],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\19.png' );
>> save('19.mat','groundTruth')
>>groundTruth{20}=struct('Boundaries',[37,38],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\20.png' );
>> save('20.mat','groundTruth')
>>groundTruth{21}=struct('Boundaries',[39,40],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\21.png' );
>> save('21.mat','groundTruth')
>>groundTruth{22}=struct('Boundaries',[41,42],'String','D:\MasterDegree\SketchTokens\BSR\BSDS500\data\images\train\22.png' );

[Stephen23]

Basically your Boundaries matrix needs to be the same size as M.

[amal alhait]

please, could you help me how to solve this error

Error using |

Matrix dimensions must agree.

Error in stTrain>stTrainTree (line 227)

M=M | M1;

Error in stTrain (line 102)

stTrainTree( opts, stream, i );

[Image Analyst]

This is just normal debugging. What are the sizes of M and M1? - just hover over them with your cursor when you stop at that line by setting a breakpoint there. Or look in the worspace panel, or (more complicated but still simple) use the "whos" command

whos M
whos M1

Debugging in MATLAB

The solution is to just make sure M and M1 match in all dimension lengths. Right now, yours don't.

[amal alhait]

These are the size of M & M1

[Image Analyst]

Yes, just as we thought. You should not try to OR those. Why do you even want to, and how do you think that can possibly work when they are not even the same size?

[Image Analyst]

Sorry, I'm not going to have time to delve into the code to debug it for you. Good luck though.

Answer 2

[Image Analyst]

If you do this:

s.groundTruth={'D:\Master Degree\SketchTokens\BSR\BSDS500\data\images\train\1.png',[100 73]};
save('1.mat','s');

Then it will save a structure s. Then when you do this:

gt=load('1.mat');

gt will be a structure with one field called s. To get s, you can do

s = gt.s;

Now, that s is a structure with one field and that field is a cell array with two cells. To get the structure you can do

ca = s.groundTruth; % Extract the cell array.

To get the string, you can do

str = ca{1}; % will be 'D:\Master Degree\SketchTokens\BSR\BSDS500\data\images\train\1.png'

and to get the array, you can do

vec = ca{2};  % Will be [100, 73]

You've chosen a very complicated, roundabout way of saving your variables.

[amal alhait]

how can I solve it please,

Error using |

Matrix dimensions must agree.

Error in stTrain>stTrainTree (line 227)

M=M | M1;

Error in stTrain (line 102)

stTrainTree( opts, stream, i );

Answer 3

[amal alhait]

Anyone can help me please to run this file without mistakes.This is my email ([e-mail]).So, I can send the images and the file