File size: 5,093 Bytes
fb8e3a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b20b64c
fb8e3a6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
# Generator model
# Critic Model

import torch
import torchvision
import torch.nn as nn
# import torch.optim as optim
from torchvision.utils import save_image
from torchvision.transforms import transforms
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')


noise_dim = 100 
img_channels = 3
gen_features = 64
critic_features = 64


class Generator(nn.Module):
  def __init__(self, noise_dim, img_channels, gen_features):
    super(Generator, self).__init__()

    self.gen = nn.Sequential(
        # Input: N x noise_dim x 1 x 1
        self._block(noise_dim, gen_features * 16, 4, 1, 0),
        self._block(gen_features * 16, gen_features * 8, 4, 2, 1),
        self._block(gen_features * 8, gen_features * 4, 4, 2, 1),
        self._block(gen_features * 4, gen_features * 2, 4, 2, 1),
        self._block(gen_features * 2, gen_features, 4, 2, 1),
        self._block(gen_features, gen_features // 2, 4, 2, 1),
        nn.ConvTranspose2d(gen_features // 2, img_channels, kernel_size=4, stride=2, padding=1),
        nn.Tanh()
        # Output: N x channels_img x 256 x 256
    )

  def _block(self, in_c, out_c, k_size, s_size, p_size): # This is a nice practice that I learned from: # https://github.com/aladdinpersson
    return nn.Sequential(
      nn.ConvTranspose2d(
          in_c, out_c,
          k_size, s_size, p_size
      ),
      nn.BatchNorm2d(out_c),
      nn.ReLU(),
    )

  def forward(self, x):
    return self.gen(x)


class Critic(nn.Module): # aka discirminator (called critic in )
    def __init__(self, img_channels, critic_features):
        super(Critic, self).__init__()
        self.critic = nn.Sequential(
            # Input: N x channels_img x 256 x 256
            nn.Conv2d(img_channels, critic_features, kernel_size=4, stride=2, padding=1),
            nn.LeakyReLU(0.2),
            self._block(critic_features, critic_features * 2, 4, 2, 1),
            self._block(critic_features * 2, critic_features * 4, 4, 2, 1),
            self._block(critic_features * 4, critic_features * 8 , 4, 2, 1),
            self._block(critic_features * 8, critic_features * 16 , 4, 2, 1),
            self._block(critic_features * 16, critic_features * 32 , 4, 2, 1),
            nn.Conv2d(critic_features * 32, 1, kernel_size=4, stride=1, padding=0)
            # Output: N x 1 x 1 x 1
        )

    def _block(self, in_c, out_c, k_size, s_size, p_size): # this is a nice practice that I learned from: # https://github.com/aladdinpersson
      return nn.Sequential(
        nn.Conv2d(
          in_c, out_c,
          k_size, s_size, p_size
        ),
        nn.BatchNorm2d(out_c),
        nn.LeakyReLU(0.2),
      )

    def forward(self, x):
        return self.critic(x)


def weights_init(m):
    if isinstance(m, (nn.Conv2d, nn.ConvTranspose2d, nn.Linear, nn.BatchNorm2d)):
        nn.init.normal_(m.weight.data, 0.0, 0.02)
        nn.init.constant_(m.bias.data, 0)

# The following gradianet penalty funciton is take from:
# https://github.com/aladdinpersson/Machine-Learning-Collection/blob/master/ML/Pytorch/GANs/4.%20WGAN-GP/utils.py 
def gradient_penalty(critic, real, fake, device):
    BATCH_SIZE, C, H, W = real.shape
    alpha = torch.rand((BATCH_SIZE, 1, 1, 1)).repeat(1, C, H, W).to(device)
    interpolated_images = real * alpha + fake * (1 - alpha)

    # calculating the critic scores
    mixed_scores = critic(interpolated_images)

    # taking the gradient of the scores w.r.t the images
    gradient = torch.autograd.grad(
        inputs=interpolated_images,
        outputs=mixed_scores,
        grad_outputs=torch.ones_like(mixed_scores),
        create_graph=True,
        retain_graph=True,
    )[0]
    
    gradient = gradient.view(gradient.shape[0], -1)
    gradient_norm = gradient.norm(2, dim=1)
    gradient_penalty = torch.mean((gradient_norm - 1) ** 2)
    return gradient_penalty


def load_model(model_type, model_path):
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    
    # Loding the model based on the model_type
    if model_type == 'generator':
        model = Generator(noise_dim, img_channels, gen_features)
    elif model_type == 'critic':
        model = Critic(img_channels, critic_features)
    else:
        raise ValueError(f"Invalid model_type: {model_type}")

    model.load_state_dict(torch.load(model_path, map_location=device))
    model.to(device)
    model.eval()

    return model

import torchvision.transforms as transforms
from PIL import Image

def generate_random_img(model):

  # Creating a random noise tensor
  noise = torch.randn(1, noise_dim, 1, 1).to(device)  # 1 is the number of images you want to generate

  # Generating an image using the trained generator
  with torch.no_grad():
      generated_image = model(noise)

  # Converting the generated tensor to a PIL image
  generated_image = generated_image.cpu().detach().squeeze(0)
  generated_image = transforms.ToPILImage()(generated_image)

  return generated_image

if __name__ == "__main__":
    model = load_model('generator','generator_model_epoch_94.pth')
    generate_random_img(model)