-
Notifications
You must be signed in to change notification settings - Fork 1
/
caae_train.py
248 lines (212 loc) · 8.01 KB
/
caae_train.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
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
236
237
238
239
240
241
242
243
244
245
246
247
248
import argparse
import os
import numpy as np
import math
import itertools
from datetime import date
import torchvision.transforms as transforms
from torchvision.utils import save_image
import torchvision
from config import *
from utils import *
from torch.utils.data import DataLoader
from torchvision import datasets
from torch.autograd import Variable
from data import Fashion_attr_prediction
from caae import * # import the model
import torch.nn as nn
import torch.nn.functional as F
import torch
# ----------
# Training
# ----------
cuda = True if torch.cuda.is_available() else False
today = date.today().strftime("%Y%m%d")
def class_noise(class_num, dim, size):
'''if class_num == CATEGORIES[0]:
mu = 3
else:
mu = -3
mean = np.ones(dim) * mu
cov = np.diag(np.ones(dim))
arr = np.random.multivariate_normal(mean, cov, size)'''
l = class_num
half = int(dim/2)
m1 = 10*np.cos((l*2*np.pi)/10)
m2 = 10*np.sin((l*2*np.pi)/10)
mean = [m1, m2]
mean = np.tile(mean, half)
v1 = [np.cos((l*2*np.pi)/10), np.sin((l*2*np.pi)/10)]
v2 = [-np.sin((l*2*np.pi)/10), np.cos((l*2*np.pi)/10)]
a1 = 8
a2 = .4
M =np.vstack((v1,v2)).T
S = np.array([[a1, 0], [0, a2]])
c = np.dot(np.dot(M, S), np.linalg.inv(M))
cov = np.zeros((dim, dim))
for i in range(half):
cov[i*2:(i+1)*2, i*2:(i+1)*2] = c
#cov = cov*cov.T
vec = np.random.multivariate_normal(mean=mean, cov=cov,
size=size)
return vec
def sample_noise(size):
noise_vector = np.zeros((size, LATENT_DIM))
'''half = int(size/2)
noise_vector[:half,:] = class_noise(CATEGORIES[0], LATENT_DIM, half)
noise_vector[half:,:] = class_noise(CATEGORIES[1], LATENT_DIM, size-half)'''
section = int(size/N_CLASSES)
for i in range(N_CLASSES):
noise_vector[i*section:min((i+1)*section, size), :] = class_noise(i, LATENT_DIM, min(section, size-section*i))
return noise_vector
def make_one_hot_real(size):
section = int(size/N_CLASSES)
indices = []
for i in range(N_CLASSES):
indices.append(range(i * section, min((i+1)*section, size)))
arr = one_hot_encode(indices, size)
return arr
def one_hot_encode(index_arr, size):
'''arr = np.zeros((len(index1) + len(index2), N_CLASSES))
arr[index1, 0] = 1
arr[index2, 1] = 1'''
arr = np.zeros((size, N_CLASSES))
for i in range(len(index_arr)):
arr[index_arr[i], i] = 1
return arr
def train(b1, b2):
# Use binary cross-entropy loss
adversarial_loss = torch.nn.BCELoss()
pixelwise_loss = torch.nn.L1Loss()
device = torch.device("cuda" if cuda else "cpu")
# Initialize generator and discriminator
encoder = Encoder().to(device)
decoder = Decoder().to(device)
discriminator = Discriminator().to(device)
if cuda:
encoder.cuda()
decoder.cuda()
discriminator.cuda()
adversarial_loss.cuda()
pixelwise_loss.cuda()
# Configure data loader
# os.makedirs("../../data/deepfashion", exist_ok=True)
dataloader = torch.utils.data.DataLoader(
Fashion_attr_prediction(
categories=CATEGORIES,
type="train",
transform=TRANSFORM_FN,
crop=True
),
batch_size=TRAIN_BATCH_SIZE,
num_workers=NUM_WORKERS,
shuffle=True,
)
print("dont loading data")
# Optimizers
optimizer_G = torch.optim.Adam(
itertools.chain(encoder.parameters(), decoder.parameters()), lr=LR, betas=(b1, b2)
)
optimizer_D = torch.optim.Adam(discriminator.parameters(), lr=LR, betas=(b1, b2))
Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
# generate fixed noise vector
n_row = 10
#fixed_noise = Variable(Tensor(np.random.normal(0, 1, (n_row ** 2, LATENT_DIM))))
#noise_vector = np.zeros((n_row**2, LATENT_DIM))
#noise_vector[:50,:] = class_noise(CATEGORIES[0], LATENT_DIM, 50)
#noise_vector[50:,:] = class_noise(CATEGORIES[1], LATENT_DIM, 50)
fixed_noise = Variable(Tensor(sample_noise(n_row**2)))
# make directory for saving images
path = "/".join([str(c) for c in [GENERATED_BASE, "caae", CONFIG_AS_STR, "train"]])
os.makedirs(path, exist_ok=True)
# save losses across all
G_losses = []
D_losses = []
#one_hot_label = one_hot_encode(range(int(TRAIN_BATCH_SIZE/2)), range(int(TRAIN_BATCH_SIZE/2), TRAIN_BATCH_SIZE))
one_hot_label = make_one_hot_real(TRAIN_BATCH_SIZE)
print("done getting hot labels")
# training loop
for epoch in range(N_EPOCHS):
for i, (imgs, labels) in enumerate(dataloader):
# Configure input
real_imgs = Variable(imgs.type(Tensor))
# Adversarial ground truths
valid = Variable(Tensor(imgs.shape[0], 1).fill_(1.0), requires_grad=False)
fake = Variable(Tensor(imgs.shape[0], 1).fill_(0.0), requires_grad=False)
# ---------------------
# Train Discriminator
# ---------------------
optimizer_D.zero_grad()
# Sample noise as discriminator ground truth
#z = Variable(Tensor(np.random.normal(0, 1, (imgs.shape[0], LATENT_DIM))))
z = Variable(Tensor(sample_noise(imgs.shape[0])))
if imgs.shape[0] == TRAIN_BATCH_SIZE:
real_labels = Variable(Tensor(one_hot_label))
else:
real_labels = Variable(Tensor(make_one_hot_real(imgs.shape[0])))
#print("made one hot labels again")
encoded_imgs = encoder(real_imgs)
indices = []
for j in range(N_CLASSES):
indices.append(np.where(labels==CATEGORIES[j])[0])
fake_labels = Variable(Tensor(one_hot_encode(indices, imgs.shape[0])))
#print("made fake labels")
# Measure discriminator's ability to classify real from generated samples
real_loss = adversarial_loss(discriminator(z, real_labels), valid)
fake_loss = adversarial_loss(discriminator(encoded_imgs.detach(), fake_labels), fake)
d_loss = 0.5 * (real_loss + fake_loss)
d_loss.backward()
optimizer_D.step()
if i % N_CRITIC == 0:
# -----------------
# Train Generator
# -----------------
optimizer_G.zero_grad()
encoded_imgs = encoder(real_imgs)
decoded_imgs = decoder(encoded_imgs)
# Loss measures generator's ability to fool the discriminator
g_loss = 0.5 * adversarial_loss(discriminator(encoded_imgs, fake_labels), valid) + 0.5 * pixelwise_loss(
decoded_imgs, real_imgs
)
g_loss.backward()
optimizer_G.step()
batches_done = epoch * len(dataloader) + i
if batches_done % 50 == 0:
print(
"[Epoch %d/%d] [Batch %d/%d] [D loss: %f] [G loss: %f]"
% (epoch, N_EPOCHS, i, len(dataloader), d_loss.item(), g_loss.item())
)
if batches_done % SAMPLE_INTERVAL == 0:
name = gen_name(today, batches_done)
if FIXED_NOISE:
sample_image(decoder=decoder, n_row=n_row, path=path, name=name, fixed_noise=fixed_noise)
else:
sample_image(decoder=decoder, n_row=n_row, path=path, name=name)
# save losses
G_losses.append(g_loss.item())
D_losses.append(d_loss.item())
if epoch % 10 == 0:
config_mid = gen_name(CATEGORIES_AS_STR, LATENT_DIM, IMG_SIZE, epoch, LR, TRAIN_BATCH_SIZE, N_CRITIC)
print("Saved Encoder to {}".format(save_model(encoder, "caae_encoder", config_mid, today)))
print("Saved Decoder to {}".format(save_model(decoder, "caae_decoder", config_mid, today)))
print("Saved Discriminator to {}".format(save_model(discriminator, "caae_discriminator", config_mid, today)))
plot_losses("caae", G_losses, D_losses, config_mid, today)
plot_losses("caae", G_losses, D_losses, CONFIG_AS_STR, today)
return encoder, decoder, discriminator
if __name__=="__main__":
os.makedirs("images", exist_ok=True)
parser = argparse.ArgumentParser()
parser.add_argument("--b1", type=float, default=0.5, help="adam: decay of first order momentum of gradient")
parser.add_argument("--b2", type=float, default=0.999, help="adam: decay of first order momentum of gradient")
parser.add_argument("--n_cpu", type=int, default=8, help="number of cpu threads to use during batch generation")
opt = parser.parse_args()
print(opt)
encoder, decoder, discriminator = train(opt.b1, opt.b2)
# ----------
# Save Model and create Training Log
# ----------
# TODO: save this to a folder logs
print(opt)
print("Saved Encoder to {}".format(save_model(encoder, "caae_encoder", CONFIG_AS_STR, today)))
print("Saved Decoder to {}".format(save_model(decoder, "caae_decoder", CONFIG_AS_STR, today)))
print("Saved Discriminator to {}".format(save_model(discriminator, "caae_discriminator", CONFIG_AS_STR, today)))