Help

[hzhXue]

Thank you for your work. I have a question. I want to change your model into a model with twice the size, but after I changed the scale in the parameter to 2, there was an error of different dimensions in the program

[claudiom4sir]

Hi @hzhXue,

It's not clear what you want to do. Do you want to perform X2 super-resolution instead of 4? Or do you want to use a bigger model?

[hzhXue]

Hi @hzhXue,

It's not clear what you want to do. Do you want to perform X2 super-resolution instead of 4? Or do you want to use a bigger model?

I want X2 super-resolution and If I enter a resolution of 960*540, the model will explode OOM, how can I change it

[claudiom4sir]

I want X2 super-resolution

The actual upscaling is performed by the VAE model, which has 2 upscaling layers. The only way for x2 upscaling is to change the VAE, but the latent space would be different, and probably the overall framework won't work well. You may try to fine-tune the denoising UNet and see what happens.

If I enter a resolution of 960*540, the model will explode OOM

Yes, the target resolution would be 3840*2160, and it would require too much memory.