add noise problem about datasetdm

[weijiawu]

thanks for your attention. This is the fundamental principle of the diffusion process, which involves adding noise first and then employing a U-Net to predict the noise.

[fubenzun]

Comparing the method of generating images and segmentation maps using a diffusion model with the method of generating images using a diffusion model and then passing the images through a Segment Anything Model, which one is more accurate

[qwe3535]

Hello, I also have a similar problem

Could you tell me if adding just one step of noise during the training phase and then generating an image allows the UNet to learn enough information?

If we add 50 steps of noise and then have the UNet predict the noise for those 50 steps to restore the image, would the trained P-decoder achieve better results?