-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_point_gan.py
169 lines (140 loc) · 5.69 KB
/
train_point_gan.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
import os
import os.path as osp
import torch
from torch.utils.data import DataLoader
from torch.optim import RMSprop
from torch.utils.tensorboard import SummaryWriter
from datasets import PointDataset
from model.point_sdf_net import PointNet, SDFGenerator
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
import numpy as np
import io
from PIL import Image
# Output Directory
output_dir = 'C:/users/migue/shapegan/outputPOINT'
os.makedirs(output_dir, exist_ok=True)
LATENT_SIZE = 128
GRADIENT_PENALITY = 10
HIDDEN_SIZE = 256
NUM_LAYERS = 8
NORM = True
device = 'cuda' if torch.cuda.is_available() else 'cpu'
G = SDFGenerator(LATENT_SIZE, HIDDEN_SIZE, NUM_LAYERS, NORM, dropout=0.0)
D = PointNet(out_channels=1)
G, D = G.to(device), D.to(device)
G_optimizer = RMSprop(G.parameters(), lr=0.0001)
D_optimizer = RMSprop(D.parameters(), lr=0.0001)
# Dataset instantiation
dataset_path = "C:/users/migue/shapegan/data/sdf"
filenames = [f for f in os.listdir(dataset_path)]
dataset = PointDataset(dataset_path, filenames)
configuration = [ # num_points, batch_size, epochs
(1024, 82, 1000),
(2048, 52, 700),
(4096, 52, 500),
(8192, 24, 400),
(16384, 12, 600),
(32768, 6, 900),
]
# Convert plot to image
def plot_to_image(figure):
# Save the plot to a PNG in memory.
buf = io.BytesIO()
plt.savefig(buf, format='png')
plt.close(figure)
buf.seek(0)
# Convert PNG buffer to PIL Image
img = Image.open(buf).convert('RGB')
image_np = np.array(img)
return image_np
# Plot points
def plot_3d_points(points):
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(points[:, 0], points[:, 1], points[:, 2])
return fig
# Save generated samples
def save_generated_samples(fake_points, epoch, num_steps):
if epoch % 100 == 0:
np.save(osp.join(output_dir, f"generated_samples_{epoch}_{num_steps}.npy"), fake_points)
if epoch % 10 == 0:
fig = plot_3d_points(fake_points)
plt.savefig(osp.join(output_dir, f"generated_samples_{epoch}_{num_steps}.png"))
plt.close(fig)
# Save the model
def save_model(model, model_name):
model_path = osp.join(output_dir, model_name)
torch.save(model.state_dict(), model_path)
# TensorBoard writer
writer = SummaryWriter()
num_steps = 0
best_loss = float('inf')
for num_points, batch_size, epochs in configuration:
dataset.num_points = num_points
loader = DataLoader(dataset, batch_size, shuffle=True, num_workers=0)
for epoch in range(1, epochs + 1):
total_loss = 0
total_D_loss = 0
total_G_loss = 0
total_gp = 0
for uniform in loader:
num_steps += 1
uniform = uniform.to(device)
u_pos, u_dist = uniform[..., :3], uniform[..., 3:]
# Train the discriminator
D_optimizer.zero_grad()
z = torch.randn(uniform.size(0), LATENT_SIZE, device=device)
fake = G(u_pos, z)
out_real = D(u_pos, u_dist)
out_fake = D(u_pos, fake)
D_loss = out_fake.mean() - out_real.mean()
# Gradient penalty
alpha = torch.rand((uniform.size(0), 1, 1), device=device)
interpolated = alpha * u_dist + (1 - alpha) * fake
interpolated.requires_grad_(True)
out = D(u_pos, interpolated)
grad = torch.autograd.grad(out, interpolated, grad_outputs=torch.ones_like(out), create_graph=True, retain_graph=True, only_inputs=True)[0]
grad_norm = grad.view(grad.size(0), -1).norm(dim=-1, p=2)
gp = GRADIENT_PENALITY * ((grad_norm - 1).pow(2).mean())
# Final discriminator loss
loss = D_loss + gp
loss.backward()
D_optimizer.step()
total_loss += loss.item()
total_D_loss += D_loss.item()
total_gp += gp.item()
# Train the generator every 5 steps
if num_steps % 5 == 0:
G_optimizer.zero_grad()
z = torch.randn(uniform.size(0), LATENT_SIZE, device=device)
fake = G(u_pos, z)
out_fake = D(u_pos, fake)
G_loss = -out_fake.mean()
G_loss.backward()
G_optimizer.step()
total_G_loss += G_loss.item()
avg_loss = total_loss / len(loader)
avg_d_loss = total_D_loss / len(loader)
avg_g_loss = total_G_loss / len(loader)
avg_gp = total_gp / len(loader)
print('Num points: {}, Epoch: {:03d}, Total Loss: {:.6f}, D Loss: {:.6f}, G Loss: {:.6f}, GP: {:.6f}'.format(
num_points, epoch, avg_loss, avg_d_loss, avg_g_loss, avg_gp))
writer.add_scalar('Total Loss', avg_loss, num_steps)
writer.add_scalar('Discriminator Loss', avg_d_loss, num_steps)
writer.add_scalar('Generator Loss', avg_g_loss, num_steps)
writer.add_scalar('Gradient Penalty', avg_gp, num_steps)
if avg_loss < best_loss:
save_model(G, "best_generator.pt")
save_model(D, "best_discriminator.pt")
best_loss = avg_loss
# visualize every 50 epochs
if epoch % 50 == 0:
with torch.no_grad():
fake_points = fake.detach().cpu().squeeze().numpy()
save_generated_samples(fake_points, epoch, num_steps)
fig = plot_3d_points(fake_points)
img = plot_to_image(fig)
img_tensor = torch.from_numpy(img).permute(2, 0, 1).float() / 255.0
writer.add_image('Generated Samples', img_tensor, num_steps)
writer.close()