-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
262 lines (222 loc) · 10.2 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
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
'''
**************************************************
@File ：AttitudeRecognition -> train
@IDE ：PyCharm
@Author ：TheOnlyMan
@Date ：2023/4/20 10:41
**************************************************
'''
import os
import random
import sys
import numpy as np
import torch
from matplotlib import pyplot as plt
from torch import nn
from torch.utils.data import DataLoader
from tqdm import tqdm
from arguments import init
from cpn.network import cpn
from dataset import DataSet
from utils.image_utils import to_numpy
from utils.model_utils import accuracy, get_keypoints_batch, switch, checkpoint
from utils.os_utils import newdir
dirname = os.path.dirname(__file__)
sys.path.append(os.path.abspath(os.path.join(dirname, 'cpn')))
seed = 3407
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
torch.cuda.manual_seed_all(seed)
# torch.autograd.set_detect_anomaly(True)
def ohkm(loss, top_k):
ohkm_loss = 0.
for i in range(loss.size()[0]):
sub_loss = loss[i]
topk_val, topk_idx = torch.topk(sub_loss, k=top_k, dim=0, sorted=False)
tmp_loss = torch.gather(sub_loss, 0, topk_idx)
ohkm_loss += torch.sum(tmp_loss) / top_k
ohkm_loss /= loss.size()[0]
return ohkm_loss
def loader(args):
train_dataset = DataSet(args, 'train')
valid_dataset = DataSet(args, 'valid')
train_loader = DataLoader(dataset=train_dataset, batch_size=args.batch_size, shuffle=True)
valid_loader = DataLoader(dataset=valid_dataset, batch_size=args.batch_size, shuffle=True)
model = cpn((64, 48), args).to(args.device)
criterion1 = nn.MSELoss().to(args.device)
criterion2 = nn.MSELoss(reduction='none').to(args.device)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=1e-5)
return train_loader, valid_loader, model, criterion1, criterion2, optimizer
def train(args, train_loader, model, criterion1, criterion2, optimizer):
model.train()
train_epoch_loss = []
global_epoch_loss = []
refine_epoch_loss = []
num_keypoints, acc_keypoints = 0, 0
for inputs, targets, origin_keypoints in tqdm(train_loader):
inputs = inputs.to(args.device)
target15, target11, target9, target7 = targets
origin_keypoints = origin_keypoints.to(args.device)
global_outputs, refine_outputs = model(inputs)
global_loss = 0.
for global_output, label in zip(global_outputs, targets):
num_points = global_output.size()[1]
global_label = label * (origin_keypoints[:, :, 2] > 1.1).type(torch.FloatTensor).view(-1, num_points, 1, 1)
global_loss += criterion1(global_output, global_label.to(args.device)) / 2.0
refine_loss = criterion2(refine_outputs, target7.to(args.device))
refine_loss = refine_loss.mean(dim=3).mean(dim=2)
zero_loss = (refine_loss * torch.Tensor(origin_keypoints[:, :, 2] < 0.1)).sum() / num_points
refine_loss *= torch.Tensor(origin_keypoints[:, :, 2] > 0.1)
refine_loss = ohkm(refine_loss, 8)
loss = global_loss + refine_loss
if args.zloss:
loss += zero_loss
global_epoch_loss.append(global_loss.item())
refine_epoch_loss.append(refine_loss.item())
train_epoch_loss.append(loss.item())
optimizer.zero_grad()
loss.backward()
optimizer.step()
keypoints = get_keypoints_batch(refine_outputs, None)
acc_keypoints += accuracy(keypoints, to_numpy(origin_keypoints), threshold=args.threshold, lim=args.lim)
num_keypoints += inputs.shape[0] * 17
return np.average(global_epoch_loss), np.average(refine_epoch_loss), \
np.average(train_epoch_loss), acc_keypoints / num_keypoints
def valid(args, valid_loader, model, criterion1, criterion2):
with torch.no_grad():
model.eval()
valid_epoch_loss = []
global_epoch_loss = []
refine_epoch_loss = []
num_keypoints, acc_keypoints = 0, 0
for inputs, inputs_flip, targets, origin_keypoints in tqdm(valid_loader):
inputs = inputs.to(args.device)
target15, target11, target9, target7 = targets
origin_keypoints = origin_keypoints.to(args.device)
global_outputs, refine_outputs = model(inputs)
global_outputs_flip, refine_outputs_flip = None, None
if not args.noflip:
inputs_flip = inputs_flip.to(args.device)
global_outputs_flip, refine_outputs_flip = model(inputs_flip)
global_outputs_flip = torch.flip(global_outputs_flip, dims=[4])
refine_outputs_flip = torch.flip(refine_outputs_flip, dims=[3])
global_outputs += switch(global_outputs_flip)
refine_outputs += switch(refine_outputs_flip)
global_outputs /= 2
refine_outputs /= 2
global_loss = 0.
for global_output, label in zip(global_outputs, targets):
num_points = global_output.size()[1]
global_label = label * (origin_keypoints[:, :, 2] > 1.1).type(torch.FloatTensor).view(-1, num_points, 1,
1)
global_loss += criterion1(global_output, global_label.to(args.device)) / 2.0
refine_loss = criterion2(refine_outputs, target7.to(args.device))
refine_loss = refine_loss.mean(dim=3).mean(dim=2)
zero_loss = (refine_loss * torch.Tensor(origin_keypoints[:, :, 2] < 0.1)).sum() / num_points
refine_loss *= torch.Tensor(origin_keypoints[:, :, 2] > 0.1)
refine_loss = ohkm(refine_loss, 8)
loss = global_loss + refine_loss
if args.zloss:
loss += zero_loss
global_epoch_loss.append(global_loss.item())
refine_epoch_loss.append(refine_loss.item())
valid_epoch_loss.append(loss.item())
keypoints = get_keypoints_batch(refine_outputs, None)
acc_keypoints += accuracy(keypoints, to_numpy(origin_keypoints), threshold=args.threshold, lim=args.lim)
num_keypoints += inputs.shape[0] * 17
return np.average(global_epoch_loss), np.average(refine_epoch_loss), \
np.average(valid_epoch_loss), acc_keypoints / num_keypoints
def logger(**kwargs):
train_global_loss = kwargs.get('train_global_loss')
train_refine_loss = kwargs.get('train_refine_loss')
train_epochs_loss = kwargs.get('train_epochs_loss')
valid_global_loss = kwargs.get('valid_global_loss')
valid_refine_loss = kwargs.get('valid_refine_loss')
valid_epochs_loss = kwargs.get('valid_epochs_loss')
train_epochs_acc = kwargs.get('train_epochs_acc')
valid_epochs_acc = kwargs.get('valid_epochs_acc')
x1 = kwargs.get('x1')
x2 = kwargs.get('x2')
path = kwargs.get('path')
fig, axes = plt.subplots(2, 2, figsize=(12, 8))
axes[0, 0].set_title('epochs loss')
axes[0, 0].plot(x1, train_epochs_loss, '-o', label="train_epochs_loss")
axes[0, 0].plot(x2, valid_epochs_loss, '-o', label="valid_epochs_loss")
axes[0, 0].legend()
axes[0, 1].set_title('epochs acc')
axes[0, 1].plot(x1, train_epochs_acc, '-o', label="train_epochs_acc")
axes[0, 1].plot(x2, valid_epochs_acc, '-o', label="valid_epochs_acc")
axes[0, 1].legend()
axes[1, 0].set_title('train global:refine')
axes[1, 0].plot(x1, train_global_loss, '-o', label="train_global_loss")
axes[1, 0].plot(x2, train_refine_loss, '-o', label="train_refine_loss")
axes[1, 0].legend()
axes[1, 1].set_title('valid global:refine')
axes[1, 1].plot(x1, valid_global_loss, '-o', label="valid_global_loss")
axes[1, 1].plot(x2, valid_refine_loss, '-o', label="valid_refine_loss")
axes[1, 1].legend()
plt.draw()
plt.savefig(os.path.join(path, 'train_valid_compare.png'))
def main(args):
try:
train_loader, valid_loader, model, criterion1, criterion2, optimizer = loader(args)
except Exception as e:
print(e)
sys.exit(0)
train_global_loss = []
train_refine_loss = []
train_epochs_loss = []
valid_global_loss = []
valid_refine_loss = []
valid_epochs_loss = []
train_epochs_acc = []
valid_epochs_acc = []
x1 = []
x2 = []
path = newdir('data')
if not os.path.exists(path):
os.makedirs(path)
checkpoint_path = os.path.join(path, 'checkpoint')
os.makedirs(checkpoint_path)
for epoch in tqdm(range(args.epochs)):
train_global, train_refine, train_loss, train_acc = train(args, train_loader, model, criterion1,
criterion2, optimizer)
train_global_loss.append(train_global)
train_refine_loss.append(train_refine)
train_epochs_loss.append(train_loss)
train_epochs_acc.append(train_acc)
x1.append(epoch + 1)
valid_global, valid_refine, valid_loss, valid_acc = valid(args, valid_loader, model, criterion1, criterion2)
valid_global_loss.append(valid_global)
valid_refine_loss.append(valid_refine)
valid_epochs_loss.append(valid_loss)
valid_epochs_acc.append(valid_acc)
x2.append(epoch + 1)
print(f'epoch {epoch + 1} train global loss:', train_global)
print(f'epoch {epoch + 1} train refine loss:', train_refine)
print(f'epoch {epoch + 1} train loss:', train_loss)
print(f'epoch {epoch + 1} valid global loss:', valid_global)
print(f'epoch {epoch + 1} valid refine loss:', valid_refine)
print(f'epoch {epoch + 1} valid loss:', valid_loss)
print(f'epoch {epoch + 1} train acc :', train_acc * 100, '%')
print(f'epoch {epoch + 1} valid acc :', valid_acc * 100, '%')
if not args.nosave:
checkpoint(model, epoch, checkpoint_path)
logger(
train_global_loss=train_global_loss,
train_refine_loss=train_refine_loss,
train_epochs_loss=train_epochs_loss,
valid_global_loss=valid_global_loss,
valid_refine_loss=valid_refine_loss,
valid_epochs_loss=valid_epochs_loss,
train_epochs_acc=train_epochs_acc,
valid_epochs_acc=valid_epochs_acc,
x1=x1,
x2=x2,
path=path,
)
print(f'Running message was saved at {path}.')
if __name__ == '__main__':
args = init('train')
main(args)