Out of memory GPU

Hello every one,

when I run my code I have in the output of my code:

Could you help me please ? What should I do ?

my code is :

…

parfor f=1:Nb_images

im = imread(fullfile(fileFolder,char(names(f))));

[ligne, colonne]=size(im);

surface_echantillon=ligne*colonne;

[counts, ~] = imhist(im,16);

t = otsuthresh(counts);

binary = imbinarize(im,t);

bw1 = bwareaopen(binary,97);

bw6=bw1;

CC_tris=bwconncomp(bw6);

stats1=regionprops(CC_tris, 'Area', 'MajorAxisLength');

idx_tris=find ([stats1.MajorAxisLength]<300);

bw6= ismember(labelmatrix(CC_tris),idx_tris);

[L4,num]=bwlabel(bw6,8);

stats4=regionprops('table', L4, 'MajorAxisLength', 'MinorAxisLength','Area', 'Circularity');

rap2=stats4.MajorAxisLength./stats4.MajorAxisLength;

idx_4=find (rap2>5 & rap2<19 & [stats4.Area]<=700 ); % & [stats4.Area]<=600%%100 et bw7= ismember(bw6,idx_4);

bw7=bw6;

D=-bwdist(~bw6);

Ld=watershed(D);

bw2=bw6;

bw2(Ld==0)=0;

mask=imextendedmin(D,2);

D2=imimposemin(D,mask);

Ld2=watershed(D2);

bw3=gpuArray(bw6);

bw3(Ld2==0)=0;

bw4=gpuArray(bw3);

figure, imagesc(bw4),title("watershed");

[L3,num]=bwlabel(bw4,8);

stats=regionprops('table', L3, 'Area','Circularity','MajorAxisLength', 'MinorAxisLength');

elong=(stats.MajorAxisLength-stats.MinorAxisLength)./(stats.MajorAxisLength+stats.MinorAxisLength);

crit=stats.MajorAxisLength./stats.MinorAxisLength;

idx3=find([stats.Area]>=7 & [stats.Area]<550 & [stats.Circularity]>=0.75 & [elong]<0.55 & [stats.MajorAxisLength]<45 & crit<3.5);

binary=ismember(L3,idx3);

[L5,num5]=bwlabel(binary,8);

F=num5;

list_F=[list_F,F];

binary=gpuArray(binary*100);

bw=gpuArray(bw*90);

resu= gpuArray(binary+bw);

exportim=['C:\Users\MATLAB\pour vérif','\',int2str(f),'.tif'];

imwrite(resu,jet,exportim);

end

[Zohra Megnounif]

[Raymond Norris]

As you can see, your GPU device has about 5 GB of total memory. I don't have any specific suggestion, but have a few comments

• I'm assuming you're running a pool of workers; otherwise, why run the parfor loop. Therefore, multiple workers will be accessing this one (?) GPU. Will that create a bottleneck?
• How much data are your gpuArray variables (bw3, bw4, binary, bw, resu)?
• What is the core of what you want to process on the GPU? For instance

Ld=watershed(D);
bw2=bw6;
bw2(Ld==0)=0;

% ...

Ld2=watershed(D2);
bw3=gpuArray(bw6);
bw3(Ld2==0)=0;

Why is bw2 on the CPU but bw3 on the GPU? Next, look how bw4 is assigned

bw4=gpuArray(bw3);
figure, imagesc(bw4),title("watershed");
[L3,num]=bwlabel(bw4,8);

To begin with, since bw3 is an gpuArray, you could assign bw4 as such

bw4 = bw3;

But do you even need to create bw4? Couldn't you use bw3 for imagesc and bwlabel?

• If I'm reading this right, you have a bug in the following line

    binary=gpuArray(binary*100);
    bw=gpuArray(bw*90);
    resu= gpuArray(binary+bw);

I don't see where bw is first initialized. And again, since binary and bw are gpuArray, adding them together will make resu a gpuArray as well.

    binary = gpuArray(binary*100);
    bw = gpuArray(bw*90);
    resu = binary+bw;

I don't know how big binary is, but is it worth mutiplying out binary * 100 on the GPU? Or bw * 90?