ceyda

/

ddpm-ema-butterflies-64

like 0

I don't know if this is what you want, but I tried to make a demo to show the step-by-step denoising process. Hope this helps.

The code blow is modified and combined from the Diffusers's Doc:

import torch
from diffusers import DDPMPipeline
from diffusers.utils import make_image_grid
from PIL import Image
import numpy as np

pipe = DDPMPipeline.from_pretrained("ceyda/ddpm-ema-butterflies-64")
pipe.to("cuda")

scheduler = pipe.scheduler
model = pipe.unet

scheduler.set_timesteps(20)

sample_size = model.config.sample_size
noise = torch.randn((1, 3, sample_size, sample_size), device="cuda")

input = noise
images = []

for t in scheduler.timesteps:
    with torch.no_grad():
        noisy_residual = model(input, t).sample
    previous_noisy_sample = scheduler.step(noisy_residual, t, input).prev_sample
    input = previous_noisy_sample
    
    image = (input / 2 + 0.5).clamp(0, 1).squeeze()
    image = (image.permute(1, 2, 0) * 255).round().to(torch.uint8).cpu().numpy()
    image = Image.fromarray(image)
    images.append(image)

make_image_grid(images, rows=4, cols=5)