GenerativeModels

GenerativeModels copied to clipboard

NoiseSchedules cosine seems wrong and lead to division by 0

7

I wanted to use a `DDPMScheduler` with a cosine scheduling and obtained images filled with nan when sampling images. I quickly inspected the code and found that it was caused...

Leirbag-gabrieL

implement #487

zgy600

Dear Author, I hope this message finds you well. Your project has deeply impressed me with its innovative approach to image processing, and I am grateful for the insights it...

blz822

Class conditioned diffusion

6

Hi! Thanks for the great work here. When working with class-conditioned diffusion, the available inferers don't offer an input of class_label, only a concat or crossatn based context. However the...

talg2324

zgy600

Trouble with AutoencoderKL

6

Hi I have a trouble when import AutoencoderKL from generative 

khanhhoaa19

Allow users to enforce zero terminal SNR in schedulers

2

Fixes #397

marksgraham

Enforce the zero terminal SNR to schedulers

11

According to "Common Diffusion Noise Schedules and Sample Steps are Flawed" (https://arxiv.org/pdf/2305.08891.pdf), the implementation of most schedulers do not use t = T in the sampling process. We should include...

Warvito

Hi, Thank you for the amazing work on generative models. I have a question regarding the generative model for Brain Imaging generation by @Warvito (https://github.com/Warvito/generative_brain). In the paper (https://arxiv.org/abs/2209.07162), what...

andreaboscutti

Add image-to-image translation tutorial (T1 to T2)

3

Please add a tutorial illustrating how to train a model (e.g. 3d pix2pix, ..) to translate a T1 into a corresponding T2 (or CT, PS,...) image. Dataset: IXI, ...

dyollb