Newer
Older
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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
# ------------------------------------------------------------------
# _____ _ _ _
# | ___(_) __| | | ___
# | |_ | |/ _` | |/ _ \
# | _| | | (_| | | __/
# |_| |_|\__,_|_|\___| WGANGP LigthningModule
# ------------------------------------------------------------------
# Formation Introduction au Deep Learning (FIDLE)
# CNRS/MIAI - https://fidle.cnrs.fr
# ------------------------------------------------------------------
# JL Parouty (Mars 2024)
import sys
import numpy as np
import torch
import torch.nn.functional as F
import torchvision
from lightning import LightningModule
class WGANGP(LightningModule):
# -------------------------------------------------------------------------
# Init
# -------------------------------------------------------------------------
#
def __init__(
self,
data_shape = (None,None,None),
latent_dim = None,
lr = 0.0002,
b1 = 0.5,
b2 = 0.999,
batch_size = 64,
lambda_gp = 10,
generator_name = None,
discriminator_name = None,
**kwargs,
):
super().__init__()
print('\n---- GAN initialization --------------------------------------------')
# ---- Hyperparameters
#
# Enable Lightning to store all the provided arguments under the self.hparams attribute.
# These hyperparameters will also be stored within the model checkpoint.
#
self.save_hyperparameters()
print('Hyperarameters are :')
for name,value in self.hparams.items():
print(f'{name:24s} : {value}')
# ---- Because we have more than one optimizer
#
self.automatic_optimization = False
# ---- Generator/Discriminator instantiation
#
print('Submodels :')
module=sys.modules['__main__']
class_g = getattr(module, generator_name)
class_d = getattr(module, discriminator_name)
self.generator = class_g( latent_dim=latent_dim, data_shape=data_shape)
self.discriminator = class_d( latent_dim=latent_dim, data_shape=data_shape)
# ---- Validation and example data
#
self.validation_z = torch.randn(8, self.hparams.latent_dim)
self.example_input_array = torch.zeros(2, self.hparams.latent_dim)
def forward(self, z):
return self.generator(z)
def adversarial_loss(self, y_pred, y):
return F.binary_cross_entropy(y_pred, y)
def gradient_penalty(self, real_images, fake_images):
# see: https://medium.com/dejunhuang/implementing-gan-and-wgan-in-pytorch-551099afde3c
batch_size = real_images.size(0)
# ---- Create interpolate images
#
# Get a random vector : size=([batch_size])
epsilon = torch.distributions.uniform.Uniform(0, 1).sample([batch_size])
# Add dimensions to match images batch : size=([batch_size,1,1,1])
epsilon = epsilon[:, None, None, None]
# Put epsilon a the right place
epsilon = epsilon.type_as(real_images)
# Do interpolation
interpolates = epsilon * fake_images + ((1 - epsilon) * real_images)
# ---- Use autograd to compute gradient
#
# The key to making this work is including `create_graph`, this means that the computations
# in this penalty will be added to the computation graph for the loss function, so that the
# second partial derivatives will be correctly computed.
#
interpolates.requires_grad_()
pred_labels = self.discriminator.forward(interpolates)
gradients = torch.autograd.grad( inputs = interpolates,
outputs = pred_labels,
grad_outputs = torch.ones_like(pred_labels),
create_graph = True,
retain_graph = True,
only_inputs = True )[0]
grad_flat = gradients.view(batch_size, -1)
grad_norm = torch.linalg.norm(grad_flat, dim=1)
grad_penalty = (grad_norm - 1) ** 2
# gp = torch.pow(grads.norm(2, dim=1) - 1, 2).mean()
return grad_penalty
def training_step(self, batch, batch_idx):
real_imgs = batch
batch_size = batch.size(0)
lambda_gp = self.hparams.lambda_gp
optimizer_g, optimizer_d = self.optimizers()
# ---- Get some latent space vectors
# We use type_as() to make sure we initialize z on the right device (GPU/CPU).
#
z = torch.randn(batch_size, self.hparams.latent_dim)
z = z.type_as(real_imgs)
# ---- Train generator ------------------------------------------------
# Generator use optimizer #0
# We try to generate false images that could mislead the discriminator
# ---------------------------------------------------------------------
#
self.toggle_optimizer(optimizer_g)
# Get fake images
fake_imgs = self.generator.forward(z)
# Get critics
critics = self.discriminator.forward(fake_imgs)
# Loss
g_loss = -critics.mean()
# Log
self.log("g_loss", g_loss, prog_bar=True)
# Backward loss
self.manual_backward(g_loss)
optimizer_g.step()
optimizer_g.zero_grad()
self.untoggle_optimizer(optimizer_g)
# ---- Train discriminator --------------------------------------------
# Discriminator use optimizer #1
# We try to make the difference between fake images and real ones
# ---------------------------------------------------------------------
#
self.toggle_optimizer(optimizer_d)
# Get critics
critics_real = self.discriminator.forward(real_imgs)
critics_fake = self.discriminator.forward(fake_imgs.detach())
# Get gradient penalty
grad_penalty = self.gradient_penalty(real_imgs, fake_imgs.detach())
# Loss
d_loss = critics_fake.mean() - critics_real.mean() + lambda_gp*grad_penalty.mean()
# Log loss
self.log("d_loss", d_loss, prog_bar=True)
# Backward
self.manual_backward(d_loss)
optimizer_d.step()
optimizer_d.zero_grad()
self.untoggle_optimizer(optimizer_d)
def configure_optimizers(self):
lr = self.hparams.lr
b1 = self.hparams.b1
b2 = self.hparams.b2
# With a GAN, we need 2 separate optimizer.
# opt_g = torch.optim.Adam(self.generator.parameters(), lr=lr, betas=(b1, b2))
# opt_d = torch.optim.Adam(self.discriminator.parameters(), lr=lr, betas=(b1, b2),)
opt_g = torch.optim.Adam(self.generator.parameters(), lr=lr)
opt_d = torch.optim.Adam(self.discriminator.parameters(), lr=lr)
return [opt_g, opt_d], []
def on_train_epoch_end(self):
# ---- Log Graph
#
if(self.current_epoch==1):
sampleImg=torch.rand((1,28,28,1))
sampleImg=sampleImg.type_as(self.generator.model[0].weight)
self.logger.experiment.add_graph(self.discriminator,sampleImg)
# ---- Log some of these images
#
z = torch.randn(self.hparams.batch_size, self.hparams.latent_dim)
z = z.type_as(self.generator.model[0].weight)
sample_imgs = self.generator(z)
sample_imgs = sample_imgs.permute(0, 3, 1, 2) # from NHWC to NCHW
grid = torchvision.utils.make_grid(tensor=sample_imgs, nrow=12, )
self.logger.experiment.add_image(f"Generated images", grid,self.current_epoch)