-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain_gmm.py
155 lines (129 loc) · 5.33 KB
/
train_gmm.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
import os
import torch
import torch.nn as nn
import torch.utils.data as data
import torch.optim as optim
import torchvision.transforms as transforms
from tensorboardX import SummaryWriter
from models import networks, dataset
import torchvision
import argparse
device = "cuda"
from distributed import (
get_rank,
synchronize,
)
mean_candidate = [0.74112587, 0.69617281, 0.68865463]
std_candidate = [0.2941623, 0.30806473, 0.30613222]
mean_clothing = [0.73949153, 0.70635068, 0.71736564]
std_clothing = [0.34867646, 0.36374153, 0.35065262]
inv_normalize = transforms.Normalize(
mean=[-m/s for m, s in zip(mean_candidate, std_candidate)],
std=[1/s for s in std_candidate]
)
clothing_normalize = transforms.Normalize(
mean=[-m/s for m, s in zip(mean_clothing, std_clothing)],
std=[1/s for s in std_clothing]
)
parser = argparse.ArgumentParser(description="Pose with Style trainer")
parser.add_argument("--local_rank", type=int, default=0, help="local rank for distributed training")
parser.add_argument("--batchSize", type=int, default=8)
parser.add_argument("--dataroot", type=str, default="data")
parser.add_argument("--datapairs", type=str, default="train_pairs.txt")
parser.add_argument("--phase", type=str, default="train")
parser.add_argument("--beta1", type=float, default=0.5)
opt_train = parser.parse_args()
torch.distributed.init_process_group(backend="nccl", init_method="env://")
torch.cuda.set_device(opt_train.local_rank)
synchronize()
def data_sampler(dataset, shuffle, distributed):
if distributed:
return data.distributed.DistributedSampler(dataset)
if shuffle:
return data.RandomSampler(dataset)
else:
return data.SequentialSampler(dataset)
train_dataset = dataset.BaseDataset(opt_train)
sampler = data_sampler(train_dataset, shuffle=True, distributed=True)
train_dataloader = torch.utils.data.DataLoader(
train_dataset,
batch_size=opt_train.batchSize,
sampler=sampler,
shuffle=False)
if get_rank() == 0:
writer = SummaryWriter('runs/gmm')
gmm = networks.GMM_OLD(7, 3).to(device)
discriminator = networks.Discriminator(10).to(device)
optimizerG = optim.AdamW(gmm.parameters(), lr=0.0002, betas=(opt_train.beta1, 0.999))
optimizerD = optim.AdamW(discriminator.parameters(), lr=0.0002, betas=(opt_train.beta1, 0.999))
gmm = nn.parallel.DistributedDataParallel(
gmm,
device_ids=[opt_train.local_rank],
output_device=opt_train.local_rank,
broadcast_buffers=False,
find_unused_parameters=True
)
discriminator = nn.parallel.DistributedDataParallel(
discriminator,
device_ids=[opt_train.local_rank],
output_device=opt_train.local_rank,
broadcast_buffers=False
)
def discriminate(netD ,input_label, real_or_fake):
input = torch.cat([input_label, real_or_fake], dim=1)
return netD.forward(input)
sigmoid = nn.Sigmoid()
tanh = torch.nn.Tanh()
l1loss = nn.L1Loss()
step = 0
criterionVGG = networks.VGGLoss()
criterionGAN = networks.GANLoss(use_lsgan=False, tensor=torch.cuda.FloatTensor)
criterionFeat = nn.L1Loss()
if get_rank() == 0:
if not os.path.isdir('checkpoint_gmm'):
os.mkdir('checkpoint_gmm')
for epoch in range(40):
for data in train_dataloader:
mask_clothes = (data['label'] == 4).float().cuda()
in_image = data['image'].cuda()
in_edge = data['edge'].cuda()
in_color = data['color'].cuda()
in_skeleton = data['skeleton'].cuda()
pre_clothes_mask = (in_edge > 0.5).float().cuda()
clothes = in_color*pre_clothes_mask
fake_c, affine = gmm(clothes, mask_clothes, in_skeleton)
fake_c *= mask_clothes
input_pool = torch.cat([clothes, mask_clothes, in_skeleton],1)
real_pool = in_image*mask_clothes
fake_pool = fake_c
D_pool = discriminator
loss_D_fake = 0
loss_D_real = 0
loss_G_VGG = 0
pred_fake = discriminate(D_pool, input_pool.detach(), fake_pool.detach())
loss_D_fake += criterionGAN(pred_fake, False)
pred_real = discriminate(D_pool, input_pool.detach(), real_pool.detach())
loss_D_real = criterionGAN(pred_real, True)
pred_fake = D_pool.forward(torch.cat((input_pool.detach(), fake_pool.detach()), dim=1))
loss_G_GAN = criterionGAN(pred_fake, True)
loss_G_VGG += criterionVGG(fake_pool, real_pool) * 5
L1_loss = criterionFeat(fake_pool, real_pool)
L1_loss += criterionFeat(affine, real_pool)
loss_D = loss_D_fake + loss_D_real
loss_G = loss_G_GAN + loss_G_VGG + L1_loss
optimizerG.zero_grad()
loss_G.backward()
optimizerG.step()
optimizerD.zero_grad()
loss_D.backward()
optimizerD.step()
if step % 500 == 0 and get_rank() == 0:
writer.add_image('warped_garment', torchvision.utils.make_grid(inv_normalize(fake_c)), step)
writer.add_image('affine', torchvision.utils.make_grid(clothing_normalize(affine)), step)
writer.add_image('gt', torchvision.utils.make_grid(inv_normalize(real_pool)), step)
step += 1
if step % 40 == 0 and get_rank() == 0:
writer.add_scalar('generator loss', loss_G, step)
writer.add_scalar('discriminator loss', loss_D, step)
if get_rank() == 0:
torch.save(gmm.module.state_dict(), "checkpoint_gmm/gmm_"+str(epoch)+".pth")