pip install transformers

from transformers import pipeline
import torch

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
pipe = pipeline(task="image-to-image", model="caidas/swin2SR-lightweight-x2-64", device=device)

from PIL import Image
import requests

url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/cat.jpg"
image = Image.open(requests.get(url, stream=True).raw)

print(image.size)

upscaled = pipe(image)
print(upscaled.size)

from transformers import Swin2SRForImageSuperResolution, Swin2SRImageProcessor

model = Swin2SRForImageSuperResolution.from_pretrained("caidas/swin2SR-lightweight-x2-64").to(device)
processor = Swin2SRImageProcessor("caidas/swin2SR-lightweight-x2-64")

pixel_values = processor(image, return_tensors="pt").pixel_values
print(pixel_values.shape)

pixel_values = pixel_values.to(device)

import torch

with torch.no_grad():
  outputs = model(pixel_values)

outputs.reconstruction.data.shape
# torch.Size([1, 3, 880, 1072])

import numpy as np

# 压缩，移至 CPU 并裁剪值
output = outputs.reconstruction.data.squeeze().cpu().clamp_(0, 1).numpy()
# 重新排列轴
output = np.moveaxis(output, source=0, destination=-1)
# 将值恢复到像素值范围
output = (output * 255.0).round().astype(np.uint8)
Image.fromarray(output)

以图生图（图像到图像任务）¶