-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_encoders.py
156 lines (125 loc) · 5.24 KB
/
train_encoders.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
import json
import argparse
import numpy as np
from pathlib import Path
from skimage.io import imsave
import torch
import torch.nn as nn
from torch.utils.data.dataloader import DataLoader
from torchvision.transforms import ToPILImage
from piq import SSIMLoss # image quality losses
from src.dataset import SuperMarioKartDataset
from src.processing import get_transforms
from src.modeling import autoencoders
"""
ToDos:
1. Resize images
2. Add noise loss
3. Refactor code
4. Use pretrained VGG layers
5. Consider adding FOV - accuracy decreases with distance
"""
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--data_path", required=True, type=Path)
parser.add_argument("--save_path", default=None, type=Path)
parser.add_argument("--load_path", default=None, type=Path)
parser.add_argument("--config_name", required=True)
parser.add_argument("--encoder_type", required=True, type=str)
parser.add_argument("--viz_samples", default=5, type=int)
parser.add_argument("--patience", default=10, type=int)
args = parser.parse_args()
return args
def main():
args = parse_args()
config = load_json('./configs', args.config_name)
config = config[args.encoder_type]
if torch.cuda.is_available():
DEVICE = 'cuda'
else:
DEVICE = 'cpu'
print('CUDA is not available, using CPU instead...')
# ===================Create data loaders=====================
transforms = get_transforms(roi=config['roi'], grayscale=config['grayscale'])
train_set = SuperMarioKartDataset(
Path(args.data_path) / 'train_files.txt', transforms=transforms
)
valid_set = SuperMarioKartDataset(
Path(args.data_path) / 'valid_files.txt', transforms=transforms
)
train_loader = DataLoader(train_set, batch_size=config['batch_size'], num_workers=4)
valid_loader = DataLoader(train_set, batch_size=1, num_workers=4)
# ===================Export some val images=====================
if not args.save_path:
save_path = args.data_path / (args.encoder_type + '_results')
else:
save_path = args.save_path
save_path.mkdir(exist_ok=True)
# ===================Instantiate models=====================
img_shape = train_set.image_shape
if args.encoder_type == 'pca':
network = autoencoders.PCA(img_shape, config['latent_dim']).cuda()
elif args.encoder_type == 'mlp':
network = autoencoders.MLP(img_shape, config['latent_dim']).cuda()
optimizer = torch.optim.Adam(network.parameters())
ssim_loss = SSIMLoss(data_range=1.)
mse_loss = nn.MSELoss()
load_path = args.load_path
if load_path and load_path.exists():
network.load_state_dict(torch.load(load_path / "best_encoder.pt"))
best_val_loss = 999
cnt_bad_epocs = 0
for epoch in range(1, config['n_epochs'] + 1):
train_losses, valid_losses = [], []
epc_save_path = save_path / f'epoch_{epoch}'
epc_save_path.mkdir(exist_ok=True)
# ===================Training=====================
for x_t in train_loader:
x_t = x_t.cuda()
encoding, decoding = network(x_t)
step_mse = mse_loss(decoding, x_t)
loss = mse_loss(decoding, x_t) + 0.1 * ssim_loss(decoding, x_t)
train_losses.append(loss.item())
optimizer.zero_grad()
loss.backward()
optimizer.step()
# ===================Validation=====================
for i, x_t in enumerate(valid_loader):
x_t = x_t.cuda()
with torch.no_grad():
encoding, decoding = network(x_t)
step_mse = mse_loss(decoding, x_t)
step_ssim = 0.1 * ssim_loss(decoding, x_t)
loss = step_mse + step_ssim
if i < args.viz_samples: # display only the first 10 samples
input_image = np.array(ToPILImage()(x_t.cpu()[0]))
decoded_image = np.array(ToPILImage()(decoding.cpu()[0]))
merged_image = np.concatenate((input_image, decoded_image))
imsave(epc_save_path / f'result_{i}.png', merged_image)
valid_losses.append(step_mse.item())
avg_trn_loss = round(np.mean(train_losses), 4)
avg_val_loss = round(np.mean(valid_losses), 4)
print(f'Epoch - {epoch} | Avg Train MSE - {avg_trn_loss} | Avg Val MSE - {avg_val_loss}')
# ===================Checkpointing=====================
if best_val_loss > avg_val_loss:
cnt_bad_epocs = 0
best_val_loss = avg_val_loss
torch.save(network.state_dict(), save_path / 'best_encoder.pt')
# ===================Early stopping=====================
if best_val_loss < avg_val_loss:
cnt_bad_epocs += 1
if cnt_bad_epocs == args.patience:
print('Network is not improving. Stopping training...')
break
def load_json(fdir, name):
"""
Load json as python object
"""
path = Path(fdir) / "{}.json".format(name)
if not path.exists():
raise FileNotFoundError("Could not find json file: {}".format(path))
with open(path, "r") as f:
obj = json.load(f)
return obj
if __name__ == '__main__':
main()