Stable Diffusion的AutoencoderKL编码latent

Stable Diffusion

映射图像到潜空间

---

---

1.AutoencoderKL

代码如下（示例）：

from diffusers import AutoencoderKL
from diffusers import DiffusionPipeline
import torch
from PIL import Image
import torchvision
from torchvision.utils import save_image
vae= AutoencoderKL.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", subfolder="vae")

pipeline.to("cuda")
image=Image.open('testdata/brad_pitt.png').convert('RGB')
image=torchvision.transforms.functional.resize(image,256)
image=torchvision.transforms.functional.to_tensor(image).to("cuda").unsqueeze(0)
feature=vae.encode(image).latent_dist.sample()
######映射图片到潜空间
feature=feature+torch.rand(feature.shape).to("cuda")
feature = vae.decode(feature).sample

可视化图像与特征（示例）：
Input image

latent

Channel 0；1；2；3