-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
164 lines (144 loc) · 6.37 KB
/
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
import torch
from torch import nn
import numpy as np
from opt import get_opts
from einops import rearrange
# dataset
from dataset import ImageDataset
from torch.utils.data import DataLoader
# models
from models import MLP, PE, Siren
# metrics
from metrics import psnr
# optimizer
from torch.optim import Adam
from torch.optim.lr_scheduler import CosineAnnealingLR
from pytorch_lightning import LightningModule, Trainer, seed_everything
from pytorch_lightning.callbacks import ModelCheckpoint, TQDMProgressBar
from pytorch_lightning.loggers import TensorBoardLogger
# log
import wandb
class CoordMlp(LightningModule):
def __init__(self, hparams):
super().__init__()
self.save_hyperparameters(hparams)
# wandb logger
wandb.init(project="coord_mlps", entity="wenboji", name=hparams.exp_name)
wandb.config = {
"leaning_rate": hparams.lr,
"epochs": hparams.num_epochs,
"batch_size": hparams.batch_size,
"architecture": hparams.arch
}
if hparams.use_pe == 'True': # positional encoding
# joint [2**i, 0; 0, 2**i] to generate positional matrix
P = 2 * np.pi * torch.cat([torch.eye(2)*2**i for i in range(10)], 1) # (2, 2*10)
self.pe = PE(P)
if hparams.arch in ['relu', 'gaussian', 'quadratic',
'multi-quadratic', 'laplacian',
'super-gaussian', 'expsin']: # different activation function
kwargs = {'a': hparams.a, 'b': hparams.b}
act = hparams.arch
if hparams.use_pe == 'True':
n_in = self.pe.out_dim
else:
n_in = 2
self.mlp = MLP(n_in=n_in, act=act,
act_trainable=hparams.act_trainable,
**kwargs)
elif hparams.arch == 'ff': # fourier feature
P = 2 * np.pi * hparams.sc * torch.normal(torch.zeros(2, 256),
torch.ones(2, 256)) # (2, 2*10)
self.pe = PE(P)
self.mlp = MLP(n_in = self.pe.out_dim)
elif hparams.arch == 'siren': # siren
self.mlp = Siren(first_omega_0 = hparams.omega_0,
hidden_omega_0 = hparams.omega_0)
self.loss = nn.MSELoss()
def forward(self, x):
if hparams.use_pe == 'True' or hparams.arch == 'ff':
x = self.pe(x)
return self.mlp(x)
def setup(self, stage=None):
self.train_dataset = ImageDataset(hparams.image_path,
hparams.img_wh,
'train')
self.val_dataset = ImageDataset(hparams.image_path,
hparams.img_wh,
'val')
def train_dataloader(self):
return DataLoader(self.train_dataset,
shuffle=True,
num_workers=4,
batch_size=self.hparams.batch_size,
pin_memory=True)
def val_dataloader(self):
return DataLoader(self.val_dataset,
shuffle=False,
num_workers=4,
batch_size=self.hparams.batch_size,
pin_memory=True)
def configure_optimizers(self):
self.optimizer = Adam(self.mlp.parameters(), lr=self.hparams.lr)
return self.optimizer
def training_step(self, batch, batch_idx):
rgb_pred = self(batch['uv'])
loss = self.loss(rgb_pred, batch['rgb'])
psnr_ = psnr(rgb_pred, batch['rgb'])
self.log('train_loss', loss)
self.log('train_psnr', psnr_, prog_bar=True)
wandb.log({"train/loss": loss,
"train/psnr": psnr_})
if hparams.arch in ['gaussian', 'quadratic',
'multi-quadratic', 'laplacian',
'super-gaussian', 'expsin']:
for i in [1, 3, 5]: # activation layers
self.log(f'act/l{i}_a', self.mlp.net[i].a)
if hasattr(self.mlp.net[i], "b"):
self.log(f'act/l{i}_b', self.mlp.net[i].b)
return loss
def validation_step(self, batch, batch_idx):
rgb_pred = self(batch['uv'])
loss = self.loss(rgb_pred, batch['rgb'])
psnr_ = psnr(rgb_pred, batch['rgb'])
log = {'val_loss': loss,
'val_psnr': psnr_,
'rgb_gt': batch['rgb'],
'rgb_pred': rgb_pred} # (B, 3)
wandb.log({"val_loss": loss,
"val_psnr": psnr_})
return log
def validation_epoch_end(self, outputs):
mean_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
mean_psnr = torch.stack([x['val_psnr'] for x in outputs]).mean()
rgb_gt = torch.cat([x['rgb_gt'] for x in outputs])
rgb_gt = rearrange(rgb_gt, '(h w) c -> c h w',
h=hparams.img_wh[1],
w=hparams.img_wh[0])
rgb_pred = torch.cat([x['rgb_pred'] for x in outputs])
rgb_pred = rearrange(rgb_pred, '(h w) c -> c h w',
h=hparams.img_wh[1],
w=hparams.img_wh[0])
self.log('val/loss', mean_loss, prog_bar=True)
self.log('val/psnr', mean_psnr, prog_bar=True)
# log the predicted rgb every epoch
wandb.log({"val/loss": mean_loss,
"val/psnr": mean_psnr,
"rgb_gt_pred": wandb.Image(torch.stack([rgb_gt, rgb_pred])),
"step": self.global_step
})
if __name__=='__main__':
hparams = get_opts()
coord_mlp = CoordMlp(hparams)
pbar = TQDMProgressBar(refresh_rate=1)
callbacks = [pbar]
trainer = Trainer(max_epochs=hparams.num_epochs,
callbacks=callbacks,
enable_model_summary=True,
accelerator='auto',
devices=1,
num_sanity_val_steps=0, # test validation is right or false at ith step
log_every_n_steps=1,
check_val_every_n_epoch=20, # do validation behind every 20 epochs
benchmark=True)
trainer.fit(coord_mlp)