Dreambooth-Stable-Diffusion icon indicating copy to clipboard operation
Dreambooth-Stable-Diffusion copied to clipboard

What transforms are used on the dataset images when training the model?

Open ProGamerGov opened this issue 3 years ago • 0 comments

I tried to find them in the code, but it wasn't obvious which transforms were being used. I'd like to be able to modify the transforms as well.

Edit:

Looks like it used these transforms:

class FrozenClipImageEmbedder(nn.Module):
    def preprocess(self, x):
        # normalize to [0,1]
        x = kornia.geometry.resize(x, (224, 224),
                                   interpolation='bicubic',align_corners=True,
                                   antialias=self.antialias)
        x = (x + 1.) / 2.
        # renormalize according to clip
        x = kornia.enhance.normalize(x, self.mean, self.std)
        return x

Source: https://github.com/JoePenna/Dreambooth-Stable-Diffusion/blob/main/ldm/modules/encoders/modules.py#L378

ProGamerGov avatar Oct 13 '22 14:10 ProGamerGov