app.py

import torch
import torch.nn as nn

from PIL import Image

import numpy as np
import gradio as gr


class DCGAN_Generator(nn.Module):
  def __init__(self):
    super(DCGAN_Generator, self).__init__()

    self.conv1 = nn.ConvTranspose2d(100, 256, 5)
    self.bn1 = nn.BatchNorm2d(256)
    self.relu1 = nn.LeakyReLU(negative_slope=0.2)

    self.conv2 = nn.ConvTranspose2d(256, 256, 5)
    self.bn2 = nn.BatchNorm2d(256)
    self.relu2 = nn.LeakyReLU(negative_slope=0.2)

    self.conv3 = nn.ConvTranspose2d(256, 128, 4)
    self.bn3 = nn.BatchNorm2d(128)
    self.relu3 = nn.LeakyReLU(negative_slope=0.2)

    self.conv4 = nn.ConvTranspose2d(128, 64, 2, 2)
    self.bn4 = nn.BatchNorm2d(64)
    self.relu4 = nn.LeakyReLU(negative_slope=0.2)

    self.conv5 = nn.ConvTranspose2d(64, 32, 3)
    self.bn5 = nn.BatchNorm2d(32)
    self.relu5 = nn.LeakyReLU(negative_slope=0.2)

    self.conv6 = nn.ConvTranspose2d(32, 1, 3)
    self.tanh1 = nn.Tanh()

  def forward(self, x):
    x = self.relu1(self.bn1(self.conv1(x)))
    x = self.relu2(self.bn2(self.conv2(x)))
    x = self.relu3(self.bn3(self.conv3(x)))
    x = self.relu4(self.bn4(self.conv4(x)))
    x = self.relu5(self.bn5(self.conv5(x)))

    return self.tanh1(self.conv6(x))

MODEL_PATH = './gan_mnist_generator_20.pt'
model = DCGAN_Generator()
model.load_state_dict(
  torch.load(
    MODEL_PATH,
    map_location=torch.device('cpu')
  )
)

def run_generative_model(use_seed="False", seed=42):
  if use_seed == "True":
    torch.random.manual_seed(seed)
  
  # Run generator model
  noise = torch.randn(1, 100, 1, 1)
  with torch.no_grad():
    im = model(noise).detach().cpu()

  # Process image
  im = torch.squeeze(im) # reduce dimension to get single image
  im = im * 128 + 128 # linear scaler from [-1, 1] to [0, 255]
  im = np.uint8(im)
  return Image.fromarray(im)

demo = gr.Interface(
  fn=run_generative_model,
  inputs=[
    gr.Radio(["True", "False"], value="False"),
    gr.Slider(0, 100, value=42),
  ],
  outputs="image",
)
demo.launch(share=True)