Specify Output Class For GAN Image Generator

In MATLAB's 'Train Generative Adversarial Network' example, the final step is to generate new images using the predict function and the following code.

%% Generate New Images

numObservations = 25;
ZNew = randn(numLatentInputs,numObservations,'single');
dlZNew = dlarray(ZNew,'CB');

if (executionEnvironment == "auto" && canUseGPU) || executionEnvironment == "gpu"
    dlZNew = gpuArray(dlZNew);
end

dlXGeneratedNew = predict(dlnetGenerator,dlZNew);

I = imtile(extractdata(dlXGeneratedNew));
I = rescale(I);
figure
image(I)
axis off
title("Generated Images")

The example produces a random sample of 25 images from 5 classes. I want to modify this to request one image from a specific class. For example, generate 5 images for the 'Dandilion' class. Can someone help me understand if it's possible to request specific generator outputs like this? Thanks in advance.

Answer 1

[Mahesh Taparia]

Hi

You can possibly try with conditional GAN to generate images of a particular class. For more information, you can refer this documentation of conditional GAN. You can refer here for image generation.

[Nagwa megahed]

Thanxs for your comment , can i ask you aquestion according to GAN ?

I hope to find the answer with you

how can i save the synthetic images( the generated images by GAN) in the disk space,

in other words, how can i save these images in a folder to use it in training a deep network ?

the example here , doesn't show the way

[Mahesh Taparia]

Hi

It is already answered in the community. You can refer the below links:

https://in.mathworks.com/matlabcentral/answers/344409-saving-images-onto-a-folder

https://in.mathworks.com/matlabcentral/answers/493838-how-to-save-image-in-specific-folder-in-matlab