Newer
Older
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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
import logging
import logging.handlers
log_formatter = logging.Formatter(
fmt="%(asctime)s - %(levelname)s: %(message)s", datefmt="%m/%d/%Y %I:%M:%S %p"
)
logger = logging.getLogger()
logger.setLevel(logging.INFO)
log_handler_console = logging.StreamHandler()
log_handler_console.setFormatter(log_formatter)
logger.addHandler(log_handler_console)
logfile = "train.log"
log_handler_file = logging.handlers.WatchedFileHandler(logfile)
log_handler_file.setFormatter(log_formatter)
logger.addHandler(log_handler_file)
logger.info("This is a test!")
logger.warning("The training loss is over 9000!")
logger.error("This is an error")
logger.info("The end")
args = parser.parse_args()
args.channel_mapping = dict(zip(args.channel_mapping, [0, 1, 2]))
log_formatter = logging.Formatter(
fmt="%(asctime)s - %(levelname)s: %(message)s", datefmt="%m/%d/%Y %I:%M:%S %p"
)
logger = logging.getLogger()
logger.setLevel(logging.INFO)
log_handler_console = logging.StreamHandler()
log_handler_console.setFormatter(log_formatter)
logger.addHandler(log_handler_console)
if args.logfile:
log_handler_file = logging.handlers.WatchedFileHandler(args.logfile)
log_handler_file.setFormatter(log_formatter)
logger.addHandler(log_handler_file)
# Setup snapshot directory
if os.path.exists(args.snapshot_dir):
if not os.path.isdir(args.snapshot_dir):
raise ValueError(
"Snapshot directory[%s] is not a directory!" % args.snapshot_dir
)
else:
os.mkdir(args.snapshot_dir)
random_seed = int(time.time())
logger.info(f"Seed RNG: {random_seed}")
torch.manual_seed(random_seed)
device = torch.device(args.device)
model = load_model(args.model, args.weights, device)
params = model.parameters()
optimizer = model.init_optimizer()
lr_scheduler = torch.optim.lr_scheduler.StepLR(
optimizer, args.decay_epoch, args.decay_rate
)
transforms_train = [transforms.RandomHorizontalFlip(), transforms.RandomVerticalFlip()]
datasets = {}
datasets["train"] = (
CSVDataset(
os.path.join(args.csv_dir, "training.csv"),
args.dataset_dir,
model,
augments=transforms_train,
illuminations=args.illuminations,
random_illumination=True,
)
if not args.illuminations_to_channel
else IlluminationChannelDataset(
os.path.join(args.csv_dir, "training.csv"),
args.dataset_dir,
model,
augments=transforms_train,
channel_mapping=args.channel_mapping,
)
)
datasets["val"] = (
CSVDataset(
os.path.join(args.csv_dir, "validation.csv"),
args.dataset_dir,
model,
illuminations=args.illuminations,
random_illumination=False,
)
if not args.illuminations_to_channel
else IlluminationChannelDataset(
os.path.join(args.csv_dir, "validation.csv"),
args.dataset_dir,
model,
channel_mapping=args.channel_mapping,
)
)
data_loaders = {}
for key in datasets:
data_loaders[key] = torch.utils.data.DataLoader(
dataset=datasets[key],
batch_size=args.batch_size,
shuffle=True,
pin_memory=True,
num_workers=8,
)
global_epoch = 0
def run_epoch(model, optimizer, data_loader, phase="train"):
if phase == "train":
model.train()
else:
model.eval()
running_loss = 0
since = time.time()
count = 1
for inputs, labels in data_loader:
print("Iteration {}/{}".format(count, len(data_loader)) + "\r", end="")
count += 1
inputs = inputs.to(device)
if type(labels) == torch.Tensor:
labels = labels.to(device)
elif type(labels) == list:
labels = [label.to(device) for label in labels]
optimizer.zero_grad()
with torch.set_grad_enabled(phase == "train"):
output = model(inputs)
loss = model.compute_loss(output, labels)
if phase == "train":
loss.backward()
optimizer.step()
running_loss += loss.item() * inputs.size(0)
return running_loss / len(data_loader.dataset), time.time() - since
try:
for epoch in range(1, args.epochs + 1):
if epoch == 1:
str_epoch = "Epoch {}/{}".format(0, args.epochs)
epoch_loss, epoch_time = run_epoch(
model, optimizer, data_loaders["val"], phase="val"
)
str_time = "Time {}s".format(round(epoch_time))
str_loss = "Loss {:.4f}".format(epoch_loss)
print("VAL: " + str_epoch + " " + str_loss + " " + str_time)
logger.info("VAL: " + str_epoch + " " + str_loss + " " + str_time)
epoch_loss, epoch_time = run_epoch(
model, optimizer, data_loaders["train"], phase="val"
)
str_time = "Time {}s".format(round(epoch_time))
str_loss = "Loss {:.4f}".format(epoch_loss)
print("TRAIN: " + str_epoch + " " + str_loss + " " + str_time)
logger.info("TRAIN: " + str_epoch + " " + str_loss + " " + str_time)
global_epoch = epoch
str_epoch = "Epoch {}/{}".format(epoch, args.epochs)
epoch_loss, epoch_time = run_epoch(model, optimizer, data_loaders["train"])
str_time = "Time {}s".format(round(epoch_time))
str_loss = "Loss {:.4f}".format(epoch_loss)
print("TRAIN: " + str_epoch + " " + str_loss + " " + str_time)
logger.info("TRAIN: " + str_epoch + " " + str_loss + " " + str_time)
# create snapshot
if epoch % args.snapshot_epoch == 0:
snapshot_file = f"{global_epoch:0{len(str(args.epochs))}d}.pth"
snapshot_file = os.path.join(args.snapshot_dir, snapshot_file)
print("Create Snapshot[{}]".format(snapshot_file))
logger.info("Create Snapshot[{}]".format(snapshot_file))
torch.save(model.state_dict(), snapshot_file)
# compute validation loss
if epoch % args.val_epoch == 0:
print("VAL: " + str_epoch, end="\r")
epoch_loss, epoch_time = run_epoch(
model, optimizer, data_loaders["val"], phase="val"
)
str_time = "Time {}s".format(round(epoch_time))
str_loss = "Loss {:.4f}".format(epoch_loss)
print("VAL: " + str_epoch + " " + str_loss + " " + str_time)
logger.info("VAL: " + str_epoch + " " + str_loss + " " + str_time)
lr_scheduler.step()
except KeyboardInterrupt:
snapshot_file = f"{global_epoch:0{len(str(args.epochs))}d}.pth"
snapshot_file = os.path.join(args.snapshot_dir, snapshot_file)
print("Create Snapshot[{}]".format(snapshot_file))
torch.save(model.state_dict(), snapshot_file)