Open YLBGHXTX opened 1 year ago
I think you can refer to image-to-image diffusion models, such as Palette, to modify this code base to perform image-to-image task. You might need to modify the edm training, consistency distillation, and image sampling parts altogether to make it work. A potential helpful reference is: https://github.com/crosszamirski/guided-I2I
Is it not possible to specify a corresponding image as supervisory information for the generated image during the generation process in the model code that was released? Please reply to me with relevant information. Thank you!