-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_rot_equiv.py
235 lines (200 loc) · 8.66 KB
/
train_rot_equiv.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 dataset_utils import MoNuSegDataset,split_dataset,Compose,RandomResize,RandomCrop,RandomPiRotate,\
RandomHorizontalFlip,RandomRotate,CenterCrop,ToTensor,Normalize,RandomAffine
import torch
from torchvision import models
import torch.nn as nn
from eval_train import train_rot_equiv
from argparse import ArgumentParser
import torch.utils.data as tud
import argparse
from eval_train import create_save_directory,save_hparams,compute_AJI
from model import CNN3
import yaml
from os.path import join
# CONSTANTS
### TYPE
def str2bool(v):
if isinstance(v, bool):
return v
if v.lower() in ('yes', 'true', 't', 'y', '1'):
return True
elif v.lower() in ('no', 'false', 'f', 'n', '0'):
return False
else:
raise argparse.ArgumentTypeError('Boolean value expected.')
def main():
#torch.manual_seed(42)
# ------------
# args
# ------------
parser = ArgumentParser()
# Yaml config file
parser.add_argument('--config', default='config/rot_equiv_config5.yaml', type=str,help="Yaml configuration files")
args = parser.parse_args()
#------------
# YAML
#------------
with open(args.config) as f:
arguments = yaml.load(f, Loader=yaml.FullLoader)
# Training args
auto_lr = arguments['training']['auto_lr']
learning_rate = arguments['training']['learning_rate']
scheduler = arguments['training']['scheduler']
wd = arguments['training']['wd']
moment = arguments['training']['moment']
batch_size = arguments['training']['batch_size']
benchmark = arguments['training']['benchmark']
num_classes = arguments['training']['num_classes']
n_epochs = arguments['training']['n_epochs']
#Equiv training
pi_rotate = arguments['equiv_training']['pi_rotate']
Loss = arguments['equiv_training']['Loss']
multi_task = arguments['equiv_training']['multi_task']
gamma = arguments['equiv_training']['gamma']
eval_every = arguments['equiv_training']['eval_every']
# Model args
model_n = arguments['model_config']['model']
eval_angle = arguments['model_config']['eval_angle']
pretrained = arguments['model_config']['pretrained']
aji = arguments['model_config']['aji']
# Data augmentation
rotate = arguments['data_augmentation']['rotate']
scale = arguments['data_augmentation']['scale']
size_img = arguments['data_augmentation']['size_img']
size_crop = arguments['data_augmentation']['size_crop']
angle_max = arguments['data_augmentation']['angle_max']
# Dataloader and gpu
nw = arguments['loader_gpu']['nw']
pm = arguments['loader_gpu']['pm']
gpu = arguments['loader_gpu']['gpu']
rot_cpu = arguments['loader_gpu']['rot_cpu']
# Datasets
split = arguments['dataset']['split']
split_ratio = arguments['dataset']['split_ratio']
dataroot_monuseg = arguments['dataset']['dataroot_monuseg']
entire_image = arguments['dataset']['entire_image']
target_size = arguments['dataset']['target_size']
stride = arguments['dataset']['stride']
# Save config
model_name = arguments['save_config']['model_name']
save_dir = arguments['save_config']['save_dir']
save_all_ep = arguments['save_config']['save_all_ep']
save_best = arguments['save_config']['save_best']
args = parser.parse_args()
# ------------
# device
# ------------
device = torch.device("cuda:"+str(gpu) if torch.cuda.is_available() else "cpu")
print("device used:",device)
# ------------
# model
# ------------
N_CLASSES = num_classes
if model_n.upper()=='FCN':
model = models.segmentation.fcn_resnet101(pretrained=pretrained,num_classes=N_CLASSES)
elif model_n.upper()=='DLV3':
model = models.segmentation.deeplabv3_resnet101(pretrained=pretrained,num_classes=N_CLASSES)
elif model_n.upper()=='CNN3':
print('CAREFUL! If you use the model CNN3, the input size MUST BE 51.')
model = CNN3()
else:
raise Exception('model must be "FCN" , "CNN3" or "DLV3"')
model.to(device)
# ------------
# data augmentation
# ------------
if size_img < size_crop:
raise Exception('Cannot have size of input images less than size of crop')
size_img = size_img
size_crop = size_crop
if scale:
min_size = 0.7
resize = 1.3
size_max=int(size_img*resize)
size_min = size_crop
size_img = size_min
else:
resize = 1
size_max=size_max=size_img*resize
if rotate:
transforms_train = Compose([
RandomResize(min_size=size_img,max_size=size_max),
RandomRotate(angle_max=angle_max,p_rotate=0.25,expand=True),
#RandomPiRotate(p_rotate=0.25),
CenterCrop(size_crop),
RandomHorizontalFlip(flip_prob=0.5),
RandomAffine(p=0.25,angle=40,translate=(0.25,0.5),scale=1.5,shear=(-45.0,45.0))
]
)
else:
transforms_train = Compose([
RandomResize(min_size=size_img,max_size=size_max),
CenterCrop(size_crop),
RandomHorizontalFlip(flip_prob=0.5),
RandomAffine(p=0.25,angle=40,translate=(0.25,0.5),scale=1.5,shear=(-45.0,45.0))
])
# ------------
# dataset and dataloader
# ------------
train_dataset_unsup = MoNuSegDataset(dataroot_monuseg,image_set='train',transforms=transforms_train,target_size=target_size,stride=stride)
if entire_image:
test_dataset = MoNuSegDataset(dataroot_monuseg,image_set='test',load_entire_image=True)
test_dataset_aji = MoNuSegDataset(dataroot_monuseg,image_set='test',load_entire_image=True,binary=False)
else:
test_dataset = MoNuSegDataset(dataroot_monuseg,image_set='test',target_size=target_size,stride=stride)
# Split dataset
split = split
if split==True:
train_dataset_sup = split_dataset(train_dataset_unsup,split_ratio)
else:
train_dataset_sup = train_dataset_unsup
# Multi task ?
if multi_task:
train_dataset_unsup = train_dataset_sup
# Print len datasets
print("There is",len(train_dataset_unsup),"images for equivariance training,",len(train_dataset_sup),"for supervised training and",\
len(test_dataset),"for validation")
dataloader_train_sup = torch.utils.data.DataLoader(train_dataset_sup, batch_size=batch_size,num_workers=nw,\
pin_memory=pm,shuffle=True,drop_last=True)
if entire_image:
dataloader_val = torch.utils.data.DataLoader(test_dataset,num_workers=nw,pin_memory=pm,\
batch_size=1) # Batch size set to 1 if we evaluate on the entire image (1000 x 1000 size)
dataloader_val_aji = torch.utils.data.DataLoader(test_dataset_aji,num_workers=nw,pin_memory=pm,\
batch_size=1)
else:
dataloader_val = torch.utils.data.DataLoader(test_dataset,num_workers=nw,pin_memory=pm,\
batch_size=batch_size)
# Decide which device we want to run on
# Auto lr finding
#if auto_lr==True:
# ------------
# save
# ------------
save_dir = create_save_directory(save_dir)
print('model will be saved in',save_dir)
save_hparams(arguments,save_dir)
print('PARAMETERS : ')
print(arguments)
print('-------------------------------------------------------------------')
# ------------
# training
# ------------
print('N_CLASSES',N_CLASSES)
criterion_supervised = nn.CrossEntropyLoss(ignore_index=N_CLASSES) # On ignore la classe border.
optimizer = torch.optim.SGD(model.parameters(),lr=learning_rate,momentum=moment,weight_decay=wd)
train_rot_equiv(model,n_epochs,dataloader_train_sup,train_dataset_unsup,dataloader_val,criterion_supervised,optimizer,\
scheduler=scheduler,Loss=Loss,gamma=gamma,batch_size=batch_size,save_folder=save_dir,\
model_name=model_name,benchmark=benchmark,angle_max=angle_max,pi_rotate=pi_rotate,AJI=aji,\
aji_loader=dataloader_val_aji,eval_every=eval_every,save_all_ep=save_all_ep,save_best=save_best\
,rot_cpu=rot_cpu,device=device,num_classes=N_CLASSES)
model = torch.load(join(save_dir,model_name+'.pt'),map_location=device)
l_angles = [180,210,240,270,300,330,0,30,60,90,120,150]
l_iou = []
for angle in l_angles:
test_dataset_aji = MoNuSegDataset(dataroot_monuseg,image_set='test',load_entire_image=True,binary=False,fixing_rotate=True,angle_fix=angle)
dataloader_val = torch.utils.data.DataLoader(test_dataset_aji,num_workers=nw,pin_memory=pm,\
batch_size=1)
aji,aji_mean = compute_AJI(model,dataloader_val,device,dist_factor=0.3,threshold=54,clean_prediction=False,it_bg=0,it_opening=0)
print('EVAL FOR ANGLE',angle,': AJI',aji_mean)
if __name__ == '__main__':
main()