-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
235 lines (203 loc) · 8.1 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
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
from __future__ import division
from models import *
from utils.logger import *
from utils.utils import *
from utils.datasets import *
from test import evaluate
from terminaltables import AsciiTable
import os
import sys
import time
import json
import datetime
import argparse
import torch
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision import transforms
from torch.autograd import Variable
import torch.optim as optim
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-c", "--config", type=str,
default="configs/config_twoobj.json", help="path to config file")
parser.add_argument("-v", "--verbose", default=False,
help="if print all info")
parser.add_argument("--continu", type=str, default=None,
help="if continuing training from checkpoint model")
opt = parser.parse_args()
print(opt)
config = json.load(open(opt.config))
print(config)
landm_set = ['twoobj', 'landmark', 'part2']
config['log_path'] = config['log_path'].format(config['type'])
os.makedirs(config['log_path'], exist_ok=True)
os.makedirs('output', exist_ok=True)
os.makedirs(config['checkpoint_path'], exist_ok=True)
logger = Logger(config['log_path'])
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Get data configuration
data_config = parse_data_config(
config['data_config'].format(config['type']))
train_path = data_config["train"]
valid_path = data_config["valid"]
class_names = load_classes(data_config["names"])
# Initiate model
model = make_model(config).to(device)
model.apply(weights_init_normal)
# If specified we start from checkpoint
if opt.continu:
config['pretrained_weights'] = opt.continu
if 'pretrained_weights' in config:
if config['pretrained_weights'].endswith(".pth"):
model.load_state_dict(torch.load(config['pretrained_weights']))
else:
model.load_darknet_weights(config['pretrained_weights'])
# Get dataloader
dataset = CSVDataset( train_path,
{'augment':True, 'multiscale':config['multiscale_training'],
'type':config['type']} )
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=config['batch_size'],
shuffle=True,
num_workers=config['n_cpu'],
pin_memory=True,
collate_fn=dataset.collate_fn,
)
optimizer = torch.optim.Adam(model.parameters(), lr=config['learning_rate'])
metrics = [
"grid_size",
"loss",
"xl",
"yl",
"x",
"y",
"w",
"h",
"conf",
"cls",
"cls_acc",
"recall50",
"recall75",
"precision",
"conf_obj",
"conf_noobj",
]
val_acc = []
modi = len(dataloader) // 5
with open('log.txt', 'w') as f:
f.write('')
for epoch in range(config['epochs']):
model.train()
start_time = time.time()
for batch_i, (_, imgs, targets) in enumerate(dataloader):
batches_done = len(dataloader) * epoch + batch_i
imgs = Variable(imgs.to(device))
targets = Variable(targets.to(device), requires_grad=False)
loss, outputs = model(imgs, targets)
loss.backward()
if batches_done % config['gradient_accumulations']:
# Accumulates gradient before each step
optimizer.step()
optimizer.zero_grad()
# ----------------
# Log progress
# ----------------
log_str = "\n---- [Epoch %d/%d, Batch %d/%d] ----\n" % (epoch,
config['epochs'], batch_i, len(dataloader))
log_str2 = "[Epoch %d/%d, Batch %d/%d] -" % (epoch,
config['epochs'], batch_i, len(dataloader))
metric_table = [["Metrics", *[f"YOLO Layer {i}"
for i in range(len(model.yolo_layers))]]]
# Log metrics at each YOLO layer
for i, metric in enumerate(metrics):
formats = {m: "%.6f" for m in metrics}
formats["grid_size"] = "%2d"
formats["cls_acc"] = "%.2f%%"
row_metrics = [formats[metric] % yolo.metrics.get(metric, 0)
for yolo in model.yolo_layers]
metric_table += [[metric, *row_metrics]]
# Tensorboard logging
tensorboard_log = []
for j, yolo in enumerate(model.yolo_layers):
for name, metric in yolo.metrics.items():
if name != "grid_size":
tensorboard_log += [(f"{name}_{j+1}", metric)]
tensorboard_log += [("loss", loss.item())]
logger.list_of_scalars_summary(tensorboard_log, batches_done)
log_str += AsciiTable(metric_table).table
log_str += f"\nTotal loss {loss.item()}"
log_str2 += f" Totloss {loss.item()}"
# Determine approximate time left for epoch
epoch_batches_left = len(dataloader) - (batch_i + 1)
secs=epoch_batches_left*(time.time()-start_time)/(batch_i + 1)
time_left = datetime.timedelta(seconds=secs)
log_str += f"\n---- ETA {time_left}"
log_str2 += f" - ETA {time_left}"
if batch_i % modi == 0:
if opt.verbose:
print(log_str)
else:
with open('{}_log.txt'.format(config['type']), 'a+') as f:
f.write(log_str)
f.write('\n')
print(log_str2)
# print(log_str)
model.seen += imgs.size(0)
if epoch % config['checkpoint_interval'] == 0:
torch.save(model.state_dict(),
config['checkpoint_path'] + f"yolov3_%s_%d.pth" % (
config['type'], epoch) )
# torch.save(model.state_dict(),
# f"checkpoints/yolov3_ckpt_%d.pth" % epoch)
if epoch % config['evaluation_interval'] == 0:
print("\n---- Evaluating Model ----")
# Evaluate the model on the validation set
precision, recall, AP, f1, ap_class, landm = evaluate(
model,
path=valid_path,
iou_thres=0.5,
conf_thres=0.5,
nms_thres=0.5,
img_size=config['img_size'],
batch_size=config['vbatch_size'],
type=config['type'],
)
evaluation_metrics = [
("val_precision", precision.mean()),
("val_recall", recall.mean()),
("val_mAP", AP.mean()),
("val_f1", f1.mean()),
]
if model.type in landm_set:
evaluation_metrics.append( ("landm", landm.mean()) )
logger.list_of_scalars_summary(evaluation_metrics, epoch)
val_acc.append(evaluation_metrics)
with open(config['val_metrics'].format(config['type']), 'w') as f:
f.write(str(val_acc))
# Print class APs and mAP
ap_table = [["Index", "Class name", "AP"]]
for i, c in enumerate(ap_class):
ap_table += [[c, class_names[c], "%.5f" % AP[i]]]
print(AsciiTable(ap_table).table)
print(f"---- mAP {AP.mean()}")
if model.type in landm_set:
dist5 = np.sum(landm<5.0)/len(landm)
dist10 = np.sum(landm<10.0)/len(landm)
print('landmark dists:')
print('\tunder5:', dist5)
print('\tunder10:', dist10)
print('\tavg_dist:', np.mean(landm))
print('\tmax_dist:', np.max(landm))
"""
python train.py --model_def config/yolov3-landmarks.cfg --batch_size 4
--data_config config/landmark.data
--pretrained_weights weights/darknet53.conv.74
"""
"""
python detect.py --model_def config/yolov3-landmarks.cfg --batch_size 4
--class_path data/custom/classes.names
--weights_path checkpoints/yolov3_ckpt_3.pth
--image_folder /scratch/hip_data/test_images/test_series7/
"""