-
Notifications
You must be signed in to change notification settings - Fork 1
/
train_loss.py
336 lines (293 loc) · 14.1 KB
/
train_loss.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
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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
from utils.load_dataset import *
from utils.custom_utils import *
def get_dataset(cfg_dataset):
if cfg_dataset['ensamble']['flag'] == False:
if cfg_dataset['train']['virtual'] != "" and cfg_dataset['train']['real'] == "":
print("Load only virtual dataset")
transforms = FallenPeople.train_transform()
train_df = load_data(cfg_dataset['train']['virtual'])
elif cfg_dataset['train']['virtual'] == "" and cfg_dataset['train']['real'] != "":
print("Load only real dataset")
transforms = FallenPeople.real_train_transform()
train_df = load_data(cfg_dataset['train']['real'])
# else:
# print("Load real and virtual dataset!")
test_df = load_data(cfg_dataset['test'])
if cfg_dataset['validation']['valid'] == "":
path_valid = cfg_dataset['root_train']
train, valid = split_train_valid(train_df, cfg_dataset['validation']['percentage_val'])
else:
train = train_df
path_valid = cfg_dataset['root_valid']
valid = load_data(cfg_dataset['validation']['valid'])
train_dt = FallenPeople(train, cfg_dataset['root_train'], transforms)
valid_dt = FallenPeople(valid, path_valid, FallenPeople.valid_test_transform())
test_dt = FallenPeople(test_df, cfg_dataset['root_test'], FallenPeople.valid_test_transform())
return train, train_dt, valid_dt, test_dt
else:
print("Load virtual and real dataset!")
train_df_vir = load_data(cfg_dataset['train']['virtual'])
train_df_real = load_data(cfg_dataset['train']['real'])
train_vir, valid_vir = split_train_valid(train_df_vir, cfg_dataset['validation']['percentage_val'])
valid_real = load_data(cfg_dataset['validation']['valid'])
test_df = load_data(cfg_dataset['test'])
train_dt_vir = FallenPeople(train_vir, cfg_dataset['ensamble']['train_virtual'], FallenPeople.train_transform())
train_dt_real = FallenPeople(train_df_real, cfg_dataset['ensamble']['train_real'], FallenPeople.real_train_transform())
valid_dt_vir = FallenPeople(valid_vir, cfg_dataset['ensamble']['train_virtual'], FallenPeople.valid_test_transform())
valid_dt_real = FallenPeople(valid_real, cfg_dataset['ensamble']['valid_real'], FallenPeople.valid_test_transform())
test_dt = FallenPeople(test_df, cfg_dataset['root_test'], FallenPeople.valid_test_transform())
return train_dt_vir, train_dt_real, valid_dt_vir, valid_dt_real, test_dt
def get_model(cfg):
print("Loading model pretrained on resnet50...")
if cfg['trainable_layers'] == "":
model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=cfg['pretrained'],
pretrained_backbone=cfg['pretrained_backbone'])
else:
model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=cfg['pretrained'],
pretrained_backbone=cfg['pretrained_backbone'],
trainable_backbone_layers= cfg['trainable_layers'])
num_classes = 1 + cfg['num_class'] # num_class + background
# get number of input features for the classifier
in_features = model.roi_heads.box_predictor.cls_score.in_features
# replace the pre-trained model's head with a new one
model.roi_heads.box_predictor = FastRCNNPredictor(in_features, num_classes)
if cfg['device'] == "cuda":
device = torch.device(cfg['device'] if torch.cuda.is_available() else "cpu")
else:
device = torch.device(cfg['device'])
model.to(device)
return model, device
def load_model(cfg, path):
print("Loading model pretrained over artificial dataset...")
if cfg['device'] == "cuda":
device = torch.device(cfg['device'] if torch.cuda.is_available() else "cpu")
else:
device = torch.device(cfg['device'])
# create a Faster R-CNN model without pre-trained
model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=cfg['pretrained'],
pretrained_backbone=cfg['pretrained_backbone'])
num_classes = 1 + cfg['num_class'] # num_class + background
# get number of input features for the classifier
in_features = model.roi_heads.box_predictor.cls_score.in_features
# replace the pre-trained model's head with a new one
model.roi_heads.box_predictor = FastRCNNPredictor(in_features, num_classes)
checkpoint = torch.load(path, map_location=device)
model.load_state_dict(checkpoint['model'])
model.to(device)
return model, device
def prep_train_ensamble(cfg_dataset, cfg_train):
train_v, train_r, valid_v, valid_r, test_dataset = get_dataset(cfg_dataset)
#training
train_len = max(len(train_v),len(train_r))
len_vir_t = round(train_len*float(cfg_dataset['ensamble']['split'])) #split is for virtual over real
len_real_t = train_len - len_vir_t
train_vir = Subset(train_v, torch.randperm(len(train_v))[:len_vir_t])
train_real = Subset(train_r, torch.randperm(len(train_r))[:len_real_t])
#validation
valid_len = max(len(valid_v),len(valid_r))
len_vir_v = round(valid_len*float(cfg_dataset['ensamble']['split'])) #split is for virtual over real
len_real_v = valid_len - len_vir_v
valid_vir = Subset(valid_v, torch.randperm(len(valid_v))[:len_vir_v])
valid_real = Subset(valid_r, torch.randperm(len(valid_r))[:len_real_v])
train_dataset = ConcatDataset([train_vir,train_real])
valid_dataset = ConcatDataset([valid_vir,valid_real])
def collate_fn(batch):
return tuple(zip(*batch))
train_data_loader = DataLoader(
train_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
shuffle=True,
collate_fn=collate_fn
)
valid_data_loader = DataLoader(
valid_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
shuffle=False,
collate_fn=collate_fn
)
test_data_loader = DataLoader(
test_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
shuffle=False,
collate_fn=collate_fn
)
return valid_dataset, test_dataset, train_data_loader, valid_data_loader, test_data_loader
def prep_train(cfg_dataset, cfg_train):
train, train_dataset, valid_dataset, test_dataset = get_dataset(cfg_dataset)
def collate_fn(batch):
return tuple(zip(*batch))
if cfg_train['sampler']:
weights = make_weights(train)
sampler = torch.utils.data.WeightedRandomSampler(torch.DoubleTensor(weights), int(len(weights)))
train_data_loader = DataLoader(
train_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
sampler = sampler,
collate_fn=collate_fn
)
else:
train_data_loader = DataLoader(
train_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
shuffle = False,
collate_fn=collate_fn
)
valid_data_loader = DataLoader(
valid_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
shuffle=False,
collate_fn=collate_fn
)
test_data_loader = DataLoader(
test_dataset,
batch_size=cfg_train['batch_size'],
num_workers = cfg_train['num_workers'],
shuffle=False,
collate_fn=collate_fn
)
return valid_dataset, test_dataset, train_data_loader, valid_data_loader, test_data_loader
def train(args):
# Opening configuration file
with open(args.conf_file, 'r') as stream:
try:
cfg_file = yaml.safe_load(stream)
except yaml.YAMLError as exc:
print(exc)
print("Reading the configuration from YALM file...")
# Retrieving cfg
cfg_train = cfg_file['train']
cfg_model = cfg_file['model']
cfg_dataset = cfg_file['dataset']
print("Loading dataset...")
if cfg_train['pretrained_model'] == "":
model, device = get_model(cfg_model)
else:
model, device = load_model(cfg_model, cfg_train['pretrained_model'])
print(f"Preparation to train the model in {device}...")
if cfg_dataset['ensamble']['flag'] == False:
valid_dataset, test_dataset, train_data_loader, valid_data_loader, test_data_loader = prep_train(cfg_dataset, cfg_train)
else:
valid_dataset, test_dataset, train_data_loader, valid_data_loader, test_data_loader = prep_train_ensamble(cfg_dataset, cfg_train)
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.SGD(params, lr=cfg_train['lr'], momentum=cfg_train['momentum'], weight_decay=cfg_train['weight_decay'])
if cfg_train['lr_scheduler']['flag'] == False:
lr_scheduler = None
else:
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=cfg_train['lr_scheduler']['lr_step_size'],
gamma=cfg_train['lr_scheduler']['lr_gamma'])
start_epoch = 0
itr = 1
total_train_loss = []
total_valid_loss = []
if cfg_train['checkpoint']:
print("Resuming checkpoint...")
checkpoint = torch.load(cfg_train['checkpoint'])
model.load_state_dict(checkpoint['model'])
optimizer.load_state_dict(checkpoint['optimizer'])
if lr_scheduler != None:
lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
start_epoch = checkpoint['epoch']
itr = checkpoint['itr']
total_train_loss = checkpoint['total_train_loss']
total_valid_loss = checkpoint['total_valid_loss']
print(f"From epoch: {start_epoch}")
losses_value = 0.0
f_log = open(cfg_train['log_file'], "w")
early_stopping = EarlyStopping(patience=3, verbose=True, path=cfg_train['path_saved_model'])
num_epochs = cfg_train['epochs']
print("Training...\n")
for epoch in range(start_epoch, num_epochs):
start_time = time.time()
# train ------------------------------
running_corrects = 0
model.train()
train_loss = []
for images, targets in tqdm(train_data_loader, desc=f'EPOCH [{epoch+1}]: '):
images = list(image.to(device) for image in images)
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
loss_dict = model(images, targets)
losses = sum(loss for loss in loss_dict.values())
losses_value = losses.item()
train_loss.append(losses_value)
optimizer.zero_grad()
losses.backward()
optimizer.step()
itr += 1
epoch_train_loss = np.mean(train_loss)
total_train_loss.append(epoch_train_loss)
# update the learning rate
if lr_scheduler is not None:
lr_scheduler.step()
# valid -------------------------------------
with torch.no_grad():
valid_loss = []
for images, targets in valid_data_loader:
images = list(image.to(device) for image in images)
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
loss_dict = model(images, targets)
losses = sum(loss for loss in loss_dict.values())
loss_value = losses.item()
valid_loss.append(loss_value)
epoch_valid_loss = np.mean(valid_loss)
total_valid_loss.append(epoch_valid_loss)
# print losses ------------------------------
f_log.write(f"Epoch Completed: {epoch+1}/{num_epochs}, Time: {time.time()-start_time}, "
f"Train Loss: {epoch_train_loss}, Valid Loss: {epoch_valid_loss}\n")
print(f"Epoch Completed: {epoch+1}/{num_epochs}, Time: {time.time()-start_time}, "
f"Train Loss: {epoch_train_loss}, Valid Loss: {epoch_valid_loss}")
#mAP and accuracy over validation
f_log.write("\nVALIDATION PHASE: ")
print("\nVALIDATION PHASE: ")
sys.stdout = f_log
elem = evaluate(model, valid_data_loader, device=device)
elem.summarize()
#classifier_performance(valid_dataset, model, device)
sys.stdout = original_stdout
#mAP and accuracy over testing every 2 epochs
if (epoch+1) % 3 == 0:
f_log.write(f"\nTESTING PHASE EPOCH {epoch+1}: ")
print(f"\nTESTING PHASE EPOCH {epoch+1}: ")
sys.stdout = f_log
el = evaluate(model, test_data_loader, device=device)
el.summarize()
#classifier_performance(test_dataset, model, device)
sys.stdout = original_stdout
checkpoint_dict = {
'model': model.state_dict(),
'optimizer': optimizer.state_dict(),
'lr_scheduler':
lr_scheduler.state_dict() if lr_scheduler is not None else None,
'epoch': epoch,
'itr': epoch+1,
'total_train_loss' : total_train_loss,
'total_valid_loss' : total_valid_loss
}
early_stopping(epoch_valid_loss, checkpoint_dict)
if early_stopping.early_stop:
print("Early stopping")
break
f_log.close()
print("Training completed!")
#plot valid-train loss
if len(total_train_loss) > 0 and len(total_valid_loss) > 0:
plt.figure(figsize=(8, 5))
plt.plot(total_train_loss, label="Train Loss")
plt.plot(total_valid_loss, label="Valid Loss")
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.legend()
plt.savefig(cfg_train['plot'])
plt.show()
print("Plot training/validation loss saved!")
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--conf_file', required=True, help="YAML config file path")
args = parser.parse_args()
train(args)