-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
69 lines (53 loc) · 1.93 KB
/
utils.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
import torch as th
import torchvision as tv
import torch.nn as nn
import pytorch_lightning as pl
import os
class EveryNStepsCheckpoint(pl.Callback):
def __init__(self, path, every_n_step):
self.path = os.path.join(path, 'checkpoints')
self.every_n_step = every_n_step
if not os.path.exists(self.path):
os.mkdir(self.path)
def on_batch_end(self, trainer, pl_module):
gs = trainer.global_step
if gs % self.every_n_step == 0 and gs != 0:
ckpt_path = f"{self.path}/model_{gs}.ckpt"
trainer.save_checkpoint(ckpt_path)
class UnNormalize():
def __init__(self, mean, std):
self.mean = mean
self.std = std
def __call__(self, tensor):
for t, m, s in zip(tensor, self.mean, self.std):
t.mul_(s).add_(m)
return tensor
class ToImage():
def __init__(self):
self.unnorm = UnNormalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
self.topil = tv.transforms.ToPILImage()
def __call__(self, x: th.Tensor, norm=True):
x = x.clone().detach()
if len(x.size()) == 4 and x.size()[0] == 1:
x = x[0]
elif len(x.size()) != 3:
raise ValueError('Wrongly shaped tensor.')
if norm:
x = self.unnorm(x)
return self.topil(x)
def initialize_weights(m):
if isinstance(m, (nn.Linear, nn.Conv2d, nn.ConvTranspose2d)):
# nn.init.kaiming_normal_(m.weight)
m.weight.data.normal_(0.0, 0.02)
if m.bias is not None:
m.bias.data.fill_(0)
elif isinstance(m, (nn.BatchNorm2d)):
# m.weight.data.fill_(1)
m.weight.data.normal_(1.0, 0.02)
if m.bias is not None:
m.bias.data.fill_(0)
def spectral_normalization(m):
if isinstance(m, (nn.Conv2d, nn.ConvTranspose2d)):
m = nn.utils.spectral_norm(m)
def nparams(model):
return sum([p.numel() for p in model.parameters() if p.requires_grad_])