-
Notifications
You must be signed in to change notification settings - Fork 0
/
VAE.py
183 lines (146 loc) · 5.92 KB
/
VAE.py
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
import os
import torch
import numpy as np
from torch import nn
from torch.optim import AdamW
from torch.nn import LeakyReLU, LayerNorm
from torch.utils.data import DataLoader, Dataset
from torch.utils.tensorboard import SummaryWriter
class VoxelDataset(Dataset):
def __init__(self, data_dir):
self.data_dir = data_dir
self.files = [f for f in os.listdir(data_dir) if f.endswith('.npy')]
def __len__(self):
return len(self.files)
def __getitem__(self, idx):
file_path = os.path.join(self.data_dir, self.files[idx])
data = np.load(file_path) # Load numpy array
data = torch.from_numpy(data).float() # Convert to PyTorch tensor
data = data.unsqueeze(0) # Add extra dimension for single-channel
return data
class EncoderBlock(nn.Module):
def __init__(self, in_channels, out_channels):
super(EncoderBlock, self).__init__()
self.conv = nn.Conv3d(in_channels, out_channels, kernel_size=4, stride=2, padding=1)
self.activation = LeakyReLU(0.2)
def forward(self, x):
x = self.conv(x)
self.norm = LayerNorm([x.size(1), x.size(2), x.size(3), x.size(4)]).to(x.device)
return self.activation(self.norm(x))
class DecoderBlock(nn.Module):
def __init__(self, in_channels, out_channels):
super(DecoderBlock, self).__init__()
self.conv = nn.ConvTranspose3d(in_channels, out_channels, kernel_size=4, stride=2, padding=1)
self.activation = LeakyReLU(0.2)
def forward(self, x):
x = self.conv(x)
self.norm = LayerNorm([x.size(1), x.size(2), x.size(3), x.size(4)]).to(x.device)
return self.activation(self.norm(x))
class VAE(nn.Module):
def __init__(self):
super(VAE, self).__init__()
# Encoder layers
self.enc1 = EncoderBlock(1, 64)
self.enc2 = EncoderBlock(64, 128)
self.enc3 = EncoderBlock(128, 256)
self.enc4 = EncoderBlock(256, 512)
self.enc5 = EncoderBlock(512, 1024)
# Latent layers
self.fc_mu = nn.Linear(8192, 250)
self.fc_var = nn.Linear(8192, 250)
# Decoder layers
self.dec1 = DecoderBlock(250, 512)
self.dec2 = DecoderBlock(512, 256)
self.dec3 = DecoderBlock(256, 128)
self.dec4 = DecoderBlock(128, 64)
self.dec5 = DecoderBlock(64, 32)
self.dec6 = DecoderBlock(32, 1)
def reparameterize(self, mu, logvar):
std = torch.exp(0.5*logvar)
eps = torch.randn_like(std)
return mu + eps*std
def forward(self, x):
# Encoder
x = self.enc1(x)
x = self.enc2(x)
x = self.enc3(x)
x = self.enc4(x)
x = self.enc5(x)
# Flatten encoder output
x = x.view(x.size(0), -1)
# Get latent variables
mu = self.fc_mu(x)
logvar = self.fc_var(x)
z = self.reparameterize(mu, logvar)
# Decoder
z = z.view(z.size(0), z.size(1), 1, 1, 1)
z = self.dec1(z)
z = self.dec2(z)
z = self.dec3(z)
z = self.dec4(z)
z = self.dec5(z)
reconstruction = torch.sigmoid(self.dec6(z))
return reconstruction, mu, logvar
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
num_epochs = 80000
batch_size = 100
learning_rate = 0.0002
# Create VAE
vae = VAE().to(device)
optimizer = AdamW(vae.parameters(), lr=learning_rate)
# Load Data
data_dir = 'C:\\Users\\migue\\shapegan\\data\\vox64'
dataset = VoxelDataset(data_dir)
data_loader = DataLoader(dataset, batch_size=batch_size, shuffle=True)
writer = SummaryWriter()
# Training loop
for epoch in range(num_epochs):
total_loss = 0
total_recon_loss = 0
total_kld_loss = 0
for i, real_data in enumerate(data_loader):
real_data = real_data.to(device)
# Forward pass
reconstruction, mu, logvar = vae(real_data)
# Loss
recon_loss = nn.functional.binary_cross_entropy(reconstruction, real_data, reduction='sum')
kld_loss = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp())
loss = recon_loss + kld_loss
# Backward pass and optimization
optimizer.zero_grad()
loss.backward()
optimizer.step()
# Accumulate loss components
total_loss += loss.item()
total_recon_loss += recon_loss.item()
total_kld_loss += kld_loss.item()
# Average loss components over all batches
avg_loss = total_loss / len(data_loader)
avg_recon_loss = total_recon_loss / len(data_loader)
avg_kld_loss = total_kld_loss / len(data_loader)
print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {avg_loss}, Recon Loss: {avg_recon_loss}, KLD Loss: {avg_kld_loss}')
# Log the loss and its components
writer.add_scalar('Loss/Total', avg_loss, epoch)
writer.add_scalar('Loss/Reconstruction', avg_recon_loss, epoch)
writer.add_scalar('Loss/KLD', avg_kld_loss, epoch)
# Log histograms of model parameters
for name, param in vae.named_parameters():
writer.add_histogram(name, param, epoch)
# Every 100 epochs, log generated samples
if epoch % 100 == 0:
with torch.no_grad():
# Assume z is your latent vector and get a sample from the normal distribution
z = torch.randn(batch_size, 250).to(device)
out = vae.decoder(z).cpu()
# Add the sample to TensorBoard
writer.add_images('Generated Samples', out, epoch)
# Visualize the distributions of the learned mu and logvar
writer.add_histogram('mu', mu, epoch)
writer.add_histogram('logvar', logvar, epoch)
# Save the trained model every 100 epochs
if epoch % 100 == 0:
torch.save(vae.state_dict(), f'vae_epoch_{epoch}.pth')
# Save the trained model
torch.save(vae.state_dict(), 'vae.pth')
# Close the SummaryWriter
writer.close()