forked from williamyang1991/DualStyleGAN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_sampler.py
71 lines (59 loc) · 2.71 KB
/
train_sampler.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
import numpy as np
import torch
import argparse
from model.sampler.icp import ICPTrainer
import os
class TrainOptions():
def __init__(self):
self.parser = argparse.ArgumentParser(description="Train Sampler")
self.parser.add_argument("style", type=str, help="target style type")
self.parser.add_argument("--iter", type=int, default=500000, help="iterations")
self.parser.add_argument("--model_path", type=str, default='./checkpoint/', help="path of the saved models")
self.parser.add_argument("--exstyle_path", type=str, default=None, help="path to the extrinsic style codes")
self.parser.add_argument("--model_name", type=str, default='sampler.pt', help="name of the saved model")
def parse(self):
self.opt = self.parser.parse_args()
args = vars(self.opt)
print('Load options')
for name, value in sorted(args.items()):
print('%s: %s' % (str(name), str(value)))
return self.opt
if __name__ == "__main__":
device = "cuda"
parser = TrainOptions()
args = parser.parse()
print('*'*50)
if args.exstyle_path is None:
if os.path.exists(os.path.join(args.model_path, args.style, 'refined_exstyle_code.npy')):
exstyles_dict = np.load(os.path.join(args.model_path, args.style, 'refined_exstyle_code.npy'),allow_pickle='TRUE').item()
else:
exstyles_dict = np.load(os.path.join(args.model_path, args.style, 'exstyle_code.npy'),allow_pickle='TRUE').item()
else:
exstyles_dict = np.load(args.exstyle_path,allow_pickle='TRUE').item()
exstyles = []
for k in exstyles_dict.keys():
exstyles += [torch.tensor(exstyles_dict[k])]
exstyles = torch.cat(exstyles, dim=0).reshape(-1,18*512)
# augment extrinsic style codes to about 1000 by duplicate and small jittering
W = torch.normal(exstyles.repeat(1000//exstyles.shape[0], 1), 0.05)
# color code
WC = W[:,512*7:].detach().cpu().numpy()
# style code
WS = W[:,0:512*7].detach().cpu().numpy()
print('Load extrinsic tyle codes successfully!')
# train color code sampler
icptc = ICPTrainer(WC, 128)
icptc.icp.netT = icptc.icp.netT.to(device)
icptc.train_icp(int(500000/WC.shape[0]))
# train structure code sampler
icpts = ICPTrainer(WS, 128)
icpts.icp.netT = icpts.icp.netT.to(device)
icpts.train_icp(int(500000/WS.shape[0]))
torch.save(
{
"color": icptc.icp.netT.state_dict(),
"structure": icpts.icp.netT.state_dict(),
},
f"%s/%s/%s"%(args.model_path, args.style, args.model_name),
)
print('Training done!')