Image processing using gpu array

Hi all,

I am now trying to process the image using gpu array.

I made the code so that the gpu can be used, but I confirmed that matlab is running using CPU instead of using GPU on the computer.

The code to use gpu is as follows.

 O = gather(video_ycbcr(gpuArray(video)))
 O_cpu = {};
 O_cpu{1} = O(:,:,001:149); % Can be easily extend to multiple GPU
 O_gpu{1} = gpuArray(O_cpu{1}); % Move array from CPU to GPU for acceleration
 R_gpu{1}=RainDetect(O_gpu{1},paras);
 R_cpu{1}=gather(cat(3,R_gpu{1}{:}));

 R = R_cpu{1};

What parts of this code need to be modified to use GPU properly on a computer?

Thank you.

[Walter Roberson]

RainDetect is not a Mathworks function and I do not find source for it anywhere. As outside observers we have no reason to expect that it will execute on GPU.

[nahyun lee]

Oh Sorry. RainDetect function and video_ycbcr function code is as follows.

Are there any problems with using GPU for these functions?

function [R,iter]=RainDetect(O,paras)
[m,n,t] = size(O);

O = num2cell(O,[1 2]);

R{1} = gpuArray(zeros(m,n));

B{1} = O{1}-R{1};

iter{1} = 0;

[R{2},B{2},iter{2}]=RainDetect_i(O{2},B{1},paras);

for i = 3:t
    disp(i);
    [R{i},B{i},iter{i}]=RainDetect_i(O{i},B{i-1},paras);
end

%% ===============================================================
function [y] = video_ycbcr(video)
[m,n,~,num_frame] = size(video);
y = zeros(m,n,num_frame);

y = gpuArray(y);

for i = 1:num_frame
    f_ycbcr = rgb2ycbcr(video(:,:,:,i));
	y(:,:,i)=f_ycbcr(:,:,1);
end
end

[Walter Roberson]

O = num2cell(O,[1 2])

It is not obvious to me that produces a cell on the gpu or that the entries will be on the gpu. I do not understand why you convert to cell instead of just indexing the input?

Answer 1

[Sai Bhargav Avula]

Hi,

The link attached below gives a better insight on how to use GPU for better performance.

https://www.mathworks.com/help/parallel-computing/measure-and-improve-gpu-performance.html

Hope this helps!