-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
265 lines (229 loc) · 10.4 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
263
264
import argparse
import os
import yaml
from tqdm import tqdm
from torch.utils.data import DataLoader
from torch.optim.lr_scheduler import CosineAnnealingLR
import models
import utils
from statistics import mean
import torch
import torch.distributed as dist
from torch.utils.data.sampler import SubsetRandomSampler
from Evaluator import evaluator
from losses import calc_loss
import shutil
import time
import numpy as np
import matplotlib.pyplot as plt
from loader import png_Dataset
from result_to_csv import result_to_csv
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("Training On Device: ", device)
local_rank = 0
parser = argparse.ArgumentParser()
parser.add_argument('--config', default="translandseg.yaml")
parser.add_argument('--resume', default='checkpoint', type=str, metavar='PATH',
help='path to checkpoint (default: none)')
parser.add_argument('--epochs', default=50, type=int, metavar='N',
help='total epochs')
parser.add_argument('--data_path_img',default="data\\Bijie\\image\\",
type=str, metavar='data', help='path to dataset')
parser.add_argument('--data_path_label',default="data\\Bijie\label\\",
type=str, metavar='data', help='path to dataset')
args = parser.parse_args(args=[])
with open(args.config, 'r') as f:
config = yaml.load(f, Loader=yaml.FullLoader)
def make_data_loader():
dataset = png_Dataset(args.data_path_img, args.data_path_label)
num_train = len(dataset)
indices = list(range(num_train))
split = int(np.floor(0.3 * num_train))+1
if True:
np.random.seed(42)
torch.manual_seed(42)
np.random.shuffle(indices)
train_idx, valid_idx = indices[split:], indices[:split]
train_sampler = SubsetRandomSampler(train_idx)
valid_sampler = SubsetRandomSampler(valid_idx)
train_loader = torch.utils.data.DataLoader(dataset, batch_size=2, sampler=train_sampler, shuffle=False,
num_workers=2, pin_memory=True)
valid_loader = torch.utils.data.DataLoader(dataset, batch_size=2, sampler=valid_sampler,shuffle=False,
num_workers=2, pin_memory=True)
return train_loader, valid_loader
def make_data_loaders():
train_loader,val_loader = make_data_loader()
return train_loader, val_loader
def prepare_training():
model = models.make(config['model']).cuda()
optimizer = utils.make_optimizer(
model.parameters(), config['optimizer'])
lr_scheduler = CosineAnnealingLR(optimizer, args.epochs, eta_min=config.get('lr_min'))
return model, optimizer, lr_scheduler
train_loader, val_loader = make_data_loaders()
x_list=[]
y_list=[]
for i,(x,y) in enumerate(train_loader) :
if i == 5:
break
mean=[0.485, 0.456, 0.406]
std=[0.229, 0.224, 0.225]
x1 = (x* torch.tensor(std).view(1, 3, 1, 1)) + torch.tensor(mean).view(1, 3, 1, 1)
x_list.append(x1)
y_list.append(y)
x_list = np.concatenate(x_list, axis=0)
y_list = np.concatenate(y_list, axis=0)
x_arr=np.array(x_list)
y_arr=np.array(y_list)
print(np.max(x_arr), np.min(x_arr),np.unique(y_arr))
def plot_func(data_list, n=2, camp='gray', mean=None, std=None):
for m in range(n):
fig=plt.figure(figsize=(30,8))
for i in range(10):
plt.subplot(1,10,i+1)
if 'cuda' in data_list[m].device.type:
img = data_list[m][i,:,:,:].cpu().numpy().transpose(1, 2, 0)
else:
img = data_list[m][i,:,:,:].numpy().transpose(1, 2, 0)
if camp == 'gray' and img.shape[-1] == 1:
plt.imshow(img, cmap=camp)
elif mean and std and img.shape[-1] == 3:
plt.imshow(img *std + mean)
else:
plt.imshow(img)
plt.show()
x_list = torch.tensor(x_list)
y_list = torch.tensor(y_list)
plot_func([x_list, y_list], n=2)
model, optimizer, lr_scheduler = prepare_training()
model = model.cuda()
sam_checkpoint = torch.load(config['sam_checkpoint'])
model.load_state_dict(sam_checkpoint, strict=False)
for name, para in model.named_parameters():
if "image_encoder" in name and "prompt_generator" not in name:
para.requires_grad_(False)
def training(train_loader, model, optimizer, epoch, args):
train_loss = 0.0
total_num = 0.0
model.train()
evaluator.reset()
tbar = tqdm(train_loader, desc='Training>>>>>>>')
for i, (x,y) in enumerate(tbar):
x.type(torch.cuda.FloatTensor)
y.type(torch.cuda.FloatTensor)
image, target = x.to(device), y.to(device)
optimizer.zero_grad()
output = model.infer(image)
loss = calc_loss(output, target.float())
loss.backward()
optimizer.step()
total_num += x.size(0)
train_loss += loss.data.cpu().numpy() * x.size(0)
pred = torch.sigmoid(output)
pred = torch.where(pred > 0.5, 1, 0) # 二值化
evaluator.add_batch(target, pred)
accuracy = evaluator.OverallAccuracy()
presion = evaluator.Precision() # 一个列表包含每个类的查准率
recall = evaluator.Recall() # 一个列表包含每个类的查全率
F1score = evaluator.F1Score() # 一个列表包含每个类的F1值
Iou = evaluator.IntersectionOverUnion() # 一个列表包含每个类的Iou值
FWIou = evaluator.Frequency_Weighted_Intersection_over_Union()
mIou = evaluator.MeanIntersectionOverUnion()
tbar.set_description('Training ->>>- Epoch: [%3d]/[%3d] Train loss: %.4f Train Accuracy:%.3f ' % (
epoch+1, args.epochs, train_loss / total_num, accuracy))
result_dict = {'accuracy': accuracy, 'presion': presion, 'recall': recall, 'F1score': F1score, 'Iou': Iou, 'FWIou': FWIou, 'mIou': mIou, 'loss': train_loss/total_num}
return train_loss/total_num, accuracy, result_dict
def validation(val_loader, model, epoch, args):
model.eval()
evaluator.reset()
tbar = tqdm(val_loader, desc='Validation>>>>>>>')
test_loss = 0.0
total_num = 0.0
for i, (x,y) in enumerate(tbar):
x.type(torch.cuda.FloatTensor)
y.type(torch.cuda.FloatTensor)
image, target = x.to(device), y.to(device)
with torch.no_grad():
output = model.infer(image)
loss = calc_loss(output, target.float())
test_loss += loss.data.cpu().numpy() * x.size(0)
total_num += x.size(0)
pred = torch.sigmoid(output)
pred = torch.where(pred > 0.5, 1, 0)
evaluator.add_batch(target, pred)
accuracy = evaluator.OverallAccuracy()
presion = evaluator.Precision()
recall = evaluator.Recall()
F1score = evaluator.F1Score()
Iou = evaluator.IntersectionOverUnion()
FWIou = evaluator.Frequency_Weighted_Intersection_over_Union()
mIou = evaluator.MeanIntersectionOverUnion()
tbar.set_description('Validation->>>- Epoch: [%3d]/[%3d] Valid loss: %.4f Valid Accuracy:%.3f ' % (
epoch+1, args.epochs, test_loss / total_num, accuracy))
result_dict = {'accuracy': accuracy, 'presion': presion, 'recall': recall, 'F1score': F1score, 'Iou': Iou, 'FWIou': FWIou, 'mIou': mIou, 'loss': test_loss/total_num}
return test_loss/total_num, accuracy, result_dict
def save_checkpoint(state, is_best, dir=None, filename='checkpoint.pth.tar'):
if dir:
if not os.path.exists(dir):
os.makedirs(dir)
print('File ['+ dir + '] Created successfully')
torch.save(state, dir + filename)
if is_best:
shutil.copyfile(dir + filename, 'model_best.pth.tar')
def early_stopping(valid_loss, epoch, args, valid_loss_min=[np.inf], i_valid=[0]):
if valid_loss <= valid_loss_min[-1]:
save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'optimizer' : optimizer.state_dict(),
}, is_best=False, dir = args.resume, filename='checkpoint_{}_{}.pth.tar'.format(epoch+1, args.epochs))
if round(valid_loss, 4) == round(valid_loss_min[-1], 4):
print(i_valid[0])
i_valid[0] += 1
valid_loss_min[-1] = valid_loss
if i_valid[0] == 10:
print('Early stopping')
return True
# Define Evaluator
evaluator = evaluator(2)
train_losses,train_accuaryes, test_losses, test_accuracyes, train_results, test_reselts= [], [], [], [], [], []
for epoch in range(args.epochs):
since = time.time()
# train for one epoch
tanin_loss, train_accuary, train_result = training(train_loader, model, optimizer, epoch, args)
test_loss, test_accuracy, test_reselt = validation(val_loader, model, epoch, args)
# cosine learning rate scheduler
lr_scheduler.step()
# log to lists
train_losses.append(tanin_loss)
train_accuaryes.append(train_accuary)
test_losses.append(test_loss)
test_accuracyes.append(test_accuracy)
train_results.append(train_result)
test_reselts.append(test_reselt)
# early stopping and save model
if early_stopping(test_loss, epoch, args):
break
time_elapsed = time.time() - since
print('{:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))
### 可视化训练过程的loss和accuracy
def plot_fig(train_losses, valid_losses, train_accuracyes, valid_accuracyes, outdir):
# 创建文件
if not os.path.exists(os.path.dirname(outdir)):
os.makedirs(os.path.dirname(outdir))
# print('File ['+ os.path.split(outdir)[0] + '] Created successfully')
plt.style.use("ggplot")
plt.figure(figsize=(10,6))
plt.plot(np.arange(1, 1 + args.epochs), np.array(train_losses), label="train_loss")
plt.plot(np.arange(1, 1 + args.epochs), np.array(valid_losses), label="val_loss")
plt.plot(np.arange(1, 1 + args.epochs), np.array(train_accuracyes), label="train_accuracyes")
plt.plot(np.arange(1, 1 + args.epochs), np.array(valid_accuracyes), label="valid_accuracyes")
plt.ylim(0, 1)
plt.title("Training and Validation Loss / Accuracy")
plt.xlabel("Epoch")
plt.ylabel("Loss / Accuracy")
plt.legend(loc="lower left")
plt.savefig(outdir + '_[Loss-Accuracy]_epoch.png')
plot_fig(train_losses, test_losses, train_accuaryes, test_accuracyes, args.resume + "picture/SAM_prompt2z_" + str(epoch))
result_to_csv(train_results, out_file=args.resume + 'picture/SAM_prompt2z_train_'+ str(args.epochs) + '.csv')
result_to_csv(test_reselts, out_file=args.resume + 'picture/SAM_prompt2z_valid_'+ str(args.epochs) + '.csv')