-
Notifications
You must be signed in to change notification settings - Fork 9
/
train_ATO.py
136 lines (106 loc) · 5.1 KB
/
train_ATO.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
import argparse
import torch
import torch.optim as optim
from torch.utils.data import DataLoader
import os
from os.path import join
from collections import defaultdict
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from utils import dataset, util
from model import model_ATO
# Training settings
parser = argparse.ArgumentParser(description="LFSSR All-to-One")
parser.add_argument("--lr", type=float, default=1e-4, help="Learning rate")
parser.add_argument("--step", type=int, default=250, help="Learning rate decay every n epochs")
parser.add_argument("--reduce", type=float, default=0.5, help="Learning rate decay")
parser.add_argument("--patch_size", type=int, default=64, help="Training patch size")
parser.add_argument("--batch_size", type=int, default=1, help="Training batch size")
parser.add_argument("--resume_epoch", type=int, default=0, help="resume from checkpoint epoch")
parser.add_argument("--max_epoch", type=int, default=700, help="maximum epoch for training")
parser.add_argument("--num_cp", type=int, default=20, help="Number of epochs for saving checkpoint")
parser.add_argument("--num_snapshot", type=int, default=1, help="Number of epochs for saving loss figure")
parser.add_argument("--dataset", type=str, default="all", help="Dataset for training")
parser.add_argument("--dataset_path", type=str, default="LFData/train_all.h5", help="Dataset file for training")
parser.add_argument("--angular_num", type=int, default=7, help="Size of angular dim")
parser.add_argument("--scale", type=int, default=2, help="SR factor")
parser.add_argument("--feature_num", type=int, default=64, help="number of feature channels")
parser.add_argument('--layer_num', action=util.StoreAsArray, type=int, nargs='+', help="number of layers in resBlocks")
opt = parser.parse_args()
# print(opt)
def main():
print(opt)
# Device configuration
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
torch.manual_seed(1)
# Data loader
print('===> Loading datasets')
train_set = dataset.TrainDataFromHdf5(opt.dataset_path, opt.scale, opt.patch_size, opt.angular_num)
train_loader = DataLoader(dataset=train_set, batch_size=opt.batch_size, shuffle=True)
print('loaded {} LFIs from {}'.format(len(train_loader), opt.dataset_path))
# Build model
print("===> building net")
model = model_ATO.ATONet(opt).to(device)
# optimizer and loss logger
print("===> setting optimizer")
optimizer = optim.Adam(filter(lambda p: p.requires_grad, model.parameters()), lr=opt.lr)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=opt.step, gamma=opt.reduce)
losslogger = defaultdict(list)
# model dir
model_dir = 'checkpoint_ATO_{}x'.format(opt.scale)
if not os.path.exists(model_dir):
os.makedirs(model_dir)
# optionally resume from a checkpoint
if opt.resume_epoch:
resume_path = join(model_dir, 'model_epoch_{}.pth'.format(opt.resume_epoch))
if os.path.isfile(resume_path):
print("==>loading checkpoint 'epoch{}'".format(resume_path))
checkpoint = torch.load(resume_path)
model.load_state_dict(checkpoint['model'])
optimizer.load_state_dict(checkpoint['optimizer'])
scheduler.load_state_dict(checkpoint['scheduler'])
losslogger = checkpoint['losslogger']
else:
print("==> no model found at 'epoch{}'".format(opt.resume_epoch))
# training
print("===> training")
for epoch in range(opt.resume_epoch + 1, opt.max_epoch):
model.train()
scheduler.step()
loss_count = 0.
for k in range(50):
for i, batch in enumerate(train_loader, 1):
view_hr = batch[0].to(device) # [N,1,h,w]
lf_lr = batch[1].to(device) # [N,an2,h//s,w//s]
ref_ind = batch[-1].to(device) # [N]
view_sr = model(lf_lr, ref_ind)
loss = L1_Charbonnier_loss(view_sr, view_hr)
loss_count += loss.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
losslogger['epoch'].append(epoch)
losslogger['loss'].append(loss_count / len(train_loader))
# checkpoint
if epoch % opt.num_cp == 0:
model_save_path = join(model_dir, "model_epoch_{}.pth".format(epoch))
state = {'epoch': epoch, 'model': model.state_dict(), 'optimizer': optimizer.state_dict(),
'scheduler': scheduler.state_dict(), 'losslogger': losslogger, }
torch.save(state, model_save_path)
print("checkpoint saved to {}".format(model_save_path))
# loss snapshot
if epoch % opt.num_snapshot == 0:
plt.figure()
plt.title('loss')
plt.plot(losslogger['epoch'], losslogger['loss'])
plt.savefig(model_dir + ".jpg")
plt.close()
def L1_Charbonnier_loss(X, Y):
eps = 1e-6
diff = torch.add(X, -Y)
error = torch.sqrt(diff * diff + eps)
loss = torch.sum(error) / torch.numel(error)
return loss
if __name__ == "__main__":
main()