-
Notifications
You must be signed in to change notification settings - Fork 8
/
train.py
180 lines (143 loc) · 6.8 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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
import os
import torch
import torch.optim as optim
from tqdm import trange
from tensorboardX import SummaryWriter
from pathlib import Path
from model import AttnTuner
from utils import calculate_loss, create_log_dir, last_checkpoint
from test import valid
from logger import Logger
from dataset import SparsePatchDataset
from settings import get_config, get_logger, print_usage
CUDA_LAUNCH_BLOCKING=2, 3
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
def train_step(step, optimizer, model, data, writer, mask=None):
corr, _, gt_E, _, _, K1s, K2s, _, _, patch1, patch2, descriptor1, descriptor2 = data[:13]
scorepatch1, scorepatch2 = (data[13].cuda(), data[14].cuda()) if len(data) > 13 else (None, None)
meanft = ((descriptor1 descriptor2) / 2.).cuda()
idx1 = 2.5 * K1s[:,:2,:2].unsqueeze(1).cuda().inverse() @ model(patch1.cuda(), scorepatch1, meanft).unsqueeze(-1)
idx1 = idx1.squeeze(-1)
idx2 = 2.5 * K2s[:,:2,:2].unsqueeze(1).cuda().inverse() @ model(patch2.cuda(), scorepatch2, meanft).unsqueeze(-1)
idx2 = idx2.squeeze(-1)
inp = corr[...,:4].cuda() torch.cat([idx1, idx2], 2) # B x N x 4
loss = calculate_loss(inp, gt_E.cuda(), K1s.cuda(), K2s.cuda(), opt.train_thr)
writer.add_scalar('train/loss', loss.item(), step 1)
optimizer.zero_grad()
loss.backward()
optimizer.step()
return [loss.item()], mask
def train(model, train_loader, valid_loader, opt, writer, logger):
optimizer = optim.Adam(model.parameters(), lr=opt.learning_rate, weight_decay=opt.weight_decay)
checkpoint_path = last_checkpoint(opt.model_path)
opt.resume = opt.resume and bool(checkpoint_path) and os.path.isfile(checkpoint_path)
if opt.resume:
logger.info('==> Resuming from checkpoint..')
checkpoint = torch.load(checkpoint_path, map_location=torch.device('cpu'))
best_acc = checkpoint['best_acc']
start_step = checkpoint['step']
model.load_state_dict(checkpoint['model'])
optimizer.load_state_dict(checkpoint['optimizer'])
logger_train = Logger(opt.log_path / 'log_train.txt', title='k2s', resume=True)
logger_valid = Logger(opt.log_path / 'log_valid.txt', title='k2s', resume=True)
else:
start_step = 0
logger_train = Logger(opt.log_path / 'log_train.txt', title='k2s')
logger_train.set_names(['Learning Rate'] ['Sampson Loss'])
logger_valid = Logger(opt.log_path / 'log_valid.txt', title='k2s')
logger_valid.set_names(['AUC5'] ['AUC10', 'AUC20'])
aucs5, aucs10, aucs20, _, _, _, loss = valid(valid_loader, model, opt, writer, start_step)
logger_valid.append([aucs5, aucs10, aucs20])
best_acc = -loss
logger.info(f"Saving initial model with va_res = {best_acc:6.3f}")
if start_step == 0:
torch.save({
'step': start_step,
'model': model.state_dict(),
'best_acc': best_acc,
'optimizer' : optimizer.state_dict(),
}, os.path.join(opt.model_path, f'checkpoint_{start_step}.pth'))
train_loader_iter = iter(train_loader)
tbar = trange(start_step, opt.train_iter)
mask = None
for step in tbar:
try:
train_data = next(train_loader_iter)
except StopIteration:
train_loader_iter = iter(train_loader)
train_data = next(train_loader_iter)
# run training
cur_lr = optimizer.param_groups[0]['lr']
loss_vals, mask = train_step(step, optimizer, model, train_data, writer, mask)
tbar.set_description('Doing: {}/{}, LR: {}, Sampson_loss: {}'\
.format(step 1, opt.train_iter, cur_lr, loss_vals[0]))
if step % 100 == 0:
logger_train.append([cur_lr] loss_vals)
# Check if we want to write validation
b_save = ((step 1) % opt.save_intv) == 0
b_validate = ((step 1) % opt.val_intv) == 0
if b_validate:
aucs5, aucs10, aucs20, _, _, _, loss = valid(valid_loader, model, opt, writer, step)
logger_valid.append([aucs5, aucs10, aucs20])
va_res = -loss
if va_res > best_acc:
logger.info(f"Saving best model with va_res = {best_acc:6.3f}")
best_acc = va_res
torch.save({
'step': step 1,
'model': model.state_dict(),
'best_acc': best_acc,
'optimizer' : optimizer.state_dict(),
}, os.path.join(opt.model_path, 'model_best.pth'))
model.train()
if b_save:
torch.save({
'step': step 1,
'model': model.state_dict(),
'best_acc': best_acc,
'optimizer' : optimizer.state_dict(),
}, os.path.join(opt.model_path, f'checkpoint_{step 1}.pth'))
if __name__ == "__main__":
# parse command line arguments
# If we have unparsed arguments, print usage and exit
opt, unparsed = get_config()
if len(unparsed) > 0:
print_usage()
exit(1)
logger = get_logger(opt)
# construct folder that should contain pre-calculated correspondences
train_data = opt.datasets.split(',') #support multiple training datasets used jointly
logger.info('Using datasets: ' ', '.join(train_data))
valid_data = train_data
train_data = ['data/' ds '/' 'train_' opt.detector '/' for ds in train_data]
valid_data = ['data/' ds '/' 'val_' opt.detector '/' for ds in valid_data]
with_score = opt.detector in ['spnn', 'splg', 'aliked']
trainset = SparsePatchDataset(train_data, opt.nfeatures, opt.fmat, not opt.sideinfo, not with_score, train=True)
valset = SparsePatchDataset(valid_data, opt.nfeatures, opt.fmat, not opt.sideinfo, not with_score)
train_loader = torch.utils.data.DataLoader(trainset, shuffle=True, num_workers=8, batch_size=opt.batchsize)
valid_loader = torch.utils.data.DataLoader(valset, shuffle=False, num_workers=8, batch_size=1)
logger.info(f"Image pairs: {len(trainset)}")
# create or load model
lentable = {
'spnn': 256,
'splg': 256,
'aliked': 128,
'dedode': 256,
'xfeat': 64
}
model = AttnTuner(lentable[opt.detector], with_score)
if len(opt.model) > 0:
model.load_state_dict(torch.load(opt.model)['model'])
model = model.cuda()
model.train()
# ----------------------------------------
logger.info(f"Starting experiment {opt.experiment}")
output_dir = Path(__file__).parent / 'results' / opt.experiment
output_dir.mkdir(exist_ok=True, parents=True)
create_log_dir(output_dir, opt)
writer = SummaryWriter(log_dir=str(output_dir))
if opt.detect_anomaly:
torch.autograd.set_detect_anomaly(True)
train(model, train_loader, valid_loader, opt, writer, logger)