denoising-diffusion-pytorch

denoising-diffusion-pytorch copied to clipboard

Hello @lucidrains Sorry if i'm mistaken, but in the line 380 of the file denoising_diffusion_pytorch.py, where the final ResNetBlock is defined for the UNet : `self.final_res_block = block_klass(dim * 2,...

MekkCyber

I'm trying to apply this model on 1d vectors. All the vectors have a starting value of 1, however, in the sampled vectors, the starting values hover around 1 but...

ZiyuanYe-afk

Train with Self-Conditoning and Generate without Gives better results

1

Hi Phil and all diffusion coder here, I am using this diffusion codes for latent text generation. After some testing, I found out if I trained my denoising model with...

Leo-T-Zang

The config of celeba-HQ

3

hello, gays I used the default parameters directly when I trained with celeba-HQ, but I found that the results were not too good, has anyone had success with celeba-HQ training?...

pokameng

DDIM sampler for Continuous time Gaussian diffusion?

5

Hello, is possible to use the DDIM sampler when the time is continuous ([continuous_time_gaussian_diffusion.py](https://github.com/lucidrains/denoising-diffusion-pytorch/blob/main/denoising_diffusion_pytorch/continuous_time_gaussian_diffusion.py))? Please can you provide a simple example of code? I try to implement it like this...

markub3327

Train on Multi-GPU

3

When I try to train this model on multi-gpu using Accelerate, an error occurs  How to solve it and is there any default Accelerate config that I can refer...

neilwen987

Dumb Question about DM vs LDM

2

Do they only differ between the use of VAE to encode the inputs into embedding (and the conditional input part)? So if I wanted to make this in latent space,...

fujistoo

Dear @lucidrains and fellow diffusion athletes, What’s the best CIFAR FID that you have been able to get? Can anyone share a successful script (hyperparameters, etc.) with FID < ~7...

je-santos

Poor performance with the denoising_diffusion_pytorch.py

6

I trained the cifar100 dataset using the trainer example code provided by the author in readme, and closed amp, image_ size=32。 However, it is still found that the sampling effect...

WindFall-wyx

Hi! So I have trained a DDPM on celebA 64x64 with the parameters described in Ho et. al's original paper. The samples generated at inference time look decent. However, I...

Sergio20f