Can we measure the difference of 2 gray scale images?

[leviethung2103]

Hello author,

Can we measure the difference of 2 gray scale images? Or this metric is only used for RGB images?

Thank you.

[leviethung2103]

I've run the model with my two gray scales. I am not sure that the value given from network is correct.

Furtheremore, the lower LPIPS is, the better image is. Am I correct. Thank you

[richzhang]

Yes, the lower means closer.

We didn't verify on perceptual judgments with grayscale images only, but I think it is reasonable to expect consistent results to color images.

[leviethung2103]

Thank you so much.

[cyrilzakka]

My apologies for re-opening this thread after such a long while - is it possible to obtain a greyscale version of vgg16_perceptual? I trained a styleGAN model on a greyscale dataset. Currently the projector uses this model to project images in latent space. Unfortunately when running the projector I get:
Dimension 1 in both shapes must be equal, but are 1 and 3. Shapes are [1,1,128,128] and [?,3,?,?]

If not, I have to retrain the GAN which I'm leaving as the last option.

[richzhang]

I think you can convert the images into 3 channels using expand