Saving GAN generated images

1

Hi,
I am following the DC GAN tutorial of pytorch for generating synthetic images. I want to store generated after the last epoch individually.
https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html
I am new to this. I don’t know how to save these images individually in a folder on local drive. Please help me.

2

Torchvision has a function called save_image(). Here is the documentation.

You might want to take a look at this old discussion to see some examples.

Here is also a minimal example.

from torchvision.utils import save_image

# Batch x Channel x Height x Width = BxCxHxW = 64x3x28x28
rand_tensor= torch.rand(64, 3,28,28) 

# Taking the first image from the batch (B)
# Img size CxHxW
img1 = rand_tensor[0]           
save_image(img1, 'img1.png')
1 Like
3

Thank you very much for your reply. This code will save only one image. Do I have to use loop to save all images individuallly?

4

Depending on what you want to save, if you want to save every image individually then yes, you can make a loop and save every one of them.

Or you can use the following code to save a grid of all of the images generated.

import torch
from torchvision.utils import save_image

t = torch.rand((64, 3, 50, 50))
save_image(t, "img_grid.png", nrow=8)

In the link provided in the previous post is the documentation for save_image and it says you can use other **kwargs specified in the make_grid function.

Hope this helps :slight_smile:

5

Thank you. It will solve my issue.

2 Likes
7

I was able to save the images individually like this:

import os

sample_dir = ‘samples’
if not os.path.exists(sample_dir):
os.makedirs(sample_dir)

if(epoch == num_epochs-1 ):
fake_fname = ‘fake-{0:0=4d}.png’.format(i)
#save_image(fake, os.path.join(sample_dir, fake_fname), nrow=8)
save_image(fake[i], os.path.join(sample_dir, fake_fname), nrow=1)

1 Like