How we can train our own dataset using stylegan model, what should be the settings? about genforce HOT 14 CLOSED

You can definitely train on your own data. For this purpose, please prepare the data (we recommend using zip format) as well as a configuration file, like this. You can customize your own hyper-parameters based on the dataset (such as image resolution, number of training samples, etc.).

from genforce.

Thanks It worked

from genforce.

from genforce.

Yes to both of these questions.

from genforce.

from genforce.

Add a setting in your configuration file. Add image_channels=1 in modules/generator/model and modules/discriminator/model. Please refer to this line. You may also need to change this line to check the image shape. If images.shape[1] == 1, repeat it for 3 times since inception model only accepts 3-channel image.

from genforce.

Could you elaborate more the changes made in inception model. As it is giving error in my case.

from genforce.

Something like

if images.shape[1] == 1:
    images = torch.tile(images, (1, 3, 1, 1))

from genforce.

After making changes in the metrics/fid.py file, It gives following error
ValueError: The input tensor should be with shape [batch_size, channel, height, width], where channel equals to 1, height, width equal to 64!
But torch.Size([32, 3, 64, 64]) is received!

from genforce.

How about setting colorful_images = torch.tile(images, (1, 3, 1, 1))

from genforce.

Still facing same valueerror issue

from genforce.

Change Line 33 to colourful_images?

from genforce.

features = inception_model(colorful_images, output_logits=False)

like this?

from genforce.

Change Line 33 to colourful_images?

It did not work for my gray scale dataset. Could you elaborate more.

from genforce.