inference.py

import torch
import torch.optim as optim
import torch.nn as nn
from tqdm import tqdm
import numpy as np
from PIL import Image
import requests
import io

from unet import Unet, ConditionalUnet

from diffusion import GaussianDiffusion, DiffusionImageAPI

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


def inference1():
  # new image from web page
  image = requests.get("https://picsum.photos/120/80").content
  return Image.open(io.BytesIO(image))

def inference(cond, x0=None, gif=False, callback=None):
  model = Unet(
    image_channels=3,
    dropout=0.1,
  )
  model = ConditionalUnet(
    unet=model,
    num_classes=13,
  )
  model.load_state_dict(torch.load("./model_final.pt", map_location=device))

  diffusion = GaussianDiffusion(
    model=model,
    noise_steps=1000,
    beta_0=1e-4,
    beta_T=0.02,
    image_size=(192, 128),
  )

  if x0 is not None:
    x0 = diffusion.normalize_image(x0)
    x0 = x0.permute(2, 0, 1)
    x0 = x0.unsqueeze(0)

  model.to(device)
  diffusion.to(device)

  imageAPI = DiffusionImageAPI(diffusion)

  new_images, versions = diffusion.sample(1,cond=cond,x0=x0, cb=callback)
  if gif:
    images = []
    for image in versions:
      images.append(imageAPI.tensor_to_image(image.squeeze(0))) 

    print(len(images))
    print(images[0])
    # make gif out of pillow images
    images[0].save('./gif_output/versions.gif',
                   save_all=True,
                   append_images=images[1:],
                   duration=100,
                   loop=0)
  return imageAPI.tensor_to_image(new_images.squeeze(0))

if __name__ == "__main__":
  inference().show()