-
Notifications
You must be signed in to change notification settings - Fork 34
/
train_refine.py
186 lines (166 loc) · 8.98 KB
/
train_refine.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
from datetime import datetime
import scipy.misc as sm
from collections import OrderedDict
import glob
import numpy as np
import socket
import timeit
# PyTorch includes
import torch
import torch.optim as optim
from torchvision import transforms
from torch.utils.data import DataLoader
# Custom includes
from dataloaders.combine_dbs import CombineDBs as combine_dbs
import dataloaders.pascal as pascal
import dataloaders.sbd as sbd
from dataloaders import custom_transforms as tr
from dataloaders.helpers import *
from networks.loss import class_cross_entropy_loss
from networks.refinementnetwork import *
from torch.nn.functional import upsample
# Set gpu_id to -1 to run in CPU mode, otherwise set the id of the corresponding gpu
gpu_id = 0
device = torch.device("cuda:"+str(gpu_id) if torch.cuda.is_available() else "cpu")
if torch.cuda.is_available():
print('Using GPU: {} '.format(gpu_id))
# Setting parameters
use_sbd = False # train with SBD
nEpochs = 100 # Number of epochs for training
resume_epoch = 0 # Default is 0, change if want to resume
p = OrderedDict() # Parameters to include in report
p['trainBatch'] = 2 # Training batch size 5
snapshot = 10 # Store a model every snapshot epochs
nInputChannels = 5 # Number of input channels (RGB + heatmap of extreme points)
p['nAveGrad'] = 1 # Average the gradient of several iterations
p['lr'] = 1e-8 # Learning rate
p['wd'] = 0.0005 # Weight decay
p['momentum'] = 0.9 # Momentum
threshold=0.95 # loss
refinement_num_max = 1 # the number of new points:
# Results and model directories (a new directory is generated for every run)
save_dir_root = os.path.join(os.path.dirname(os.path.abspath(__file__)))
exp_name = os.path.dirname(os.path.abspath(__file__)).split('/')[-1]
if resume_epoch == 0:
runs = sorted(glob.glob(os.path.join(save_dir_root, 'run_*')))
run_id = int(runs[-1].split('_')[-1]) + 1 if runs else 0
else:
run_id = 0
save_dir = os.path.join(save_dir_root, 'run_' + str(run_id))
if not os.path.exists(os.path.join(save_dir, 'models')):
os.makedirs(os.path.join(save_dir, 'models'))
# Network definition
modelName = 'IOG_pascal_refinement'
net = Network(nInputChannels=nInputChannels,num_classes=1,
backbone='resnet101',
output_stride=16,
sync_bn=None,
freeze_bn=False,
pretrained=True)
if resume_epoch == 0:
print("Initializing from pretrained model")
else:
print("Initializing weights from: {}".format(
os.path.join(save_dir, 'models', modelName + '_epoch-' + str(resume_epoch - 1) + '.pth')))
net.load_state_dict(
torch.load(os.path.join(save_dir, 'models', modelName + '_epoch-' + str(resume_epoch - 1) + '.pth'),
map_location=lambda storage, loc: storage))
train_params = [{'params': net.get_1x_lr_params(), 'lr': p['lr']},
{'params': net.get_10x_lr_params(), 'lr': p['lr'] * 10}]
net.to(device)
if resume_epoch != nEpochs:
# Logging into Tensorboard
log_dir = os.path.join(save_dir, 'models', datetime.now().strftime('%b%d_%H-%M-%S') + '_' + socket.gethostname())
# Use the following optimizer
optimizer = optim.SGD(train_params, lr=p['lr'], momentum=p['momentum'], weight_decay=p['wd'])
p['optimizer'] = str(optimizer)
# Preparation of the data loaders
composed_transforms_tr = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.ScaleNRotate(rots=(-20, 20), scales=(.75, 1.25)),
tr.CropFromMask(crop_elems=('image', 'gt','void_pixels'), relax=30, zero_pad=True),
tr.FixedResize(resolutions={'crop_image': (512, 512), 'crop_gt': (512, 512), 'crop_void_pixels': (512, 512)},flagvals={'crop_image':cv2.INTER_LINEAR,'crop_gt':cv2.INTER_LINEAR,'crop_void_pixels': cv2.INTER_LINEAR}),
tr.IOGPoints(sigma=10, elem='crop_gt',pad_pixel=10),
tr.ToImage(norm_elem='IOG_points'),
tr.ConcatInputs(elems=('crop_image', 'IOG_points')),
tr.ToTensor()])
composed_transforms_ts = transforms.Compose([
tr.CropFromMask(crop_elems=('image', 'gt','void_pixels'), relax=30, zero_pad=True),
tr.FixedResize(resolutions={'crop_image': (512, 512), 'crop_gt': (512, 512), 'crop_void_pixels': (512, 512)},flagvals={'crop_image':cv2.INTER_LINEAR,'crop_gt':cv2.INTER_LINEAR,'crop_void_pixels': cv2.INTER_LINEAR}),
tr.IOGPoints(sigma=10, elem='crop_gt',pad_pixel=10),
tr.ToImage(norm_elem='IOG_points'),
tr.ConcatInputs(elems=('crop_image', 'IOG_points')),
tr.ToTensor()])
voc_train = pascal.VOCSegmentation(split='train', transform=composed_transforms_tr)
voc_val = pascal.VOCSegmentation(split='val', transform=composed_transforms_ts)
if use_sbd:
sbd = sbd.SBDSegmentation(split=['train', 'val'], transform=composed_transforms_tr, retname=True)
db_train = combine_dbs([voc_train, sbd], excluded=[voc_val])
else:
db_train = voc_train
p['dataset_train'] = str(db_train)
p['transformations_train'] = [str(tran) for tran in composed_transforms_tr.transforms]
trainloader = DataLoader(db_train, batch_size=p['trainBatch'], shuffle=True, num_workers=2)
# Train variables
num_img_tr = len(trainloader)
running_loss_tr = 0.0
aveGrad = 0
print("Training Network")
for epoch in range(resume_epoch, nEpochs):
start_time = timeit.default_timer()
epoch_loss = []
net.train()
for ii, sample_batched in enumerate(trainloader):
gts = sample_batched['crop_gt']
inputs = sample_batched['concat']
void_pixels = sample_batched['crop_void_pixels']
IOG_points = sample_batched['IOG_points']
inputs.requires_grad_()
inputs, gts ,void_pixels,IOG_points = inputs.to(device), gts.to(device), void_pixels.to(device), IOG_points.to(device)
out = net.forward(inputs,IOG_points,gts,refinement_num_max+1)
for i in range(0,refinement_num_max+1):
glo1,glo2,glo3,glo4,refine,iou_i=out[i]
output_glo1 = upsample(glo1, size=(512, 512), mode='bilinear', align_corners=True)
output_glo2 = upsample(glo2, size=(512, 512), mode='bilinear', align_corners=True)
output_glo3 = upsample(glo3, size=(512, 512), mode='bilinear', align_corners=True)
output_glo4 = upsample(glo4, size=(512, 512), mode='bilinear', align_corners=True)
output_refine = upsample(refine, size=(512, 512), mode='bilinear', align_corners=True)
# Compute the losses, side outputs and fuse
loss_output_glo1 = class_cross_entropy_loss(output_glo1, gts, void_pixels=void_pixels,size_average=False, batch_average=True)
loss_output_glo2 = class_cross_entropy_loss(output_glo2, gts, void_pixels=void_pixels,size_average=False, batch_average=True)
loss_output_glo3 = class_cross_entropy_loss(output_glo3, gts, void_pixels=void_pixels,size_average=False, batch_average=True)
loss_output_glo4 = class_cross_entropy_loss(output_glo4, gts, void_pixels=void_pixels,size_average=False, batch_average=True)
loss_output_refine = class_cross_entropy_loss(output_refine, gts, void_pixels=void_pixels,size_average=False, batch_average=True)
if i ==0:
loss1 = loss_output_glo1+loss_output_glo2+ loss_output_glo3+loss_output_glo4+loss_output_glo4+loss_output_refine
iou1 = iou_i
if i ==1:
loss2 = loss_output_glo1+loss_output_glo2+ loss_output_glo3+loss_output_glo4+loss_output_glo4+loss_output_refine
iou2 = iou_i
if iou1>=threshold:
loss=loss1
else:
loss=0.5*loss1+0.5*loss2
if ii % 10 ==0:
print('Epoch',epoch,'step',ii,'loss',loss)
running_loss_tr += loss.item()
# Print stuff
if ii % num_img_tr == num_img_tr - 1 -p['trainBatch']:
running_loss_tr = running_loss_tr / num_img_tr
print('[Epoch: %d, numImages: %5d]' % (epoch, ii*p['trainBatch']+inputs.data.shape[0]))
print('Loss: %f' % running_loss_tr)
running_loss_tr = 0
stop_time = timeit.default_timer()
print("Execution time: " + str(stop_time - start_time)+"\n")
# Backward the averaged gradient
loss /= p['nAveGrad']
loss.backward()
aveGrad += 1
# Update the weights once in p['nAveGrad'] forward passes
if aveGrad % p['nAveGrad'] == 0:
optimizer.step()
optimizer.zero_grad()
aveGrad = 0
# Save the model
if (epoch % snapshot) == snapshot - 1 and epoch != 0:
torch.save(net.state_dict(), os.path.join(save_dir, 'models', modelName + '_epoch-' + str(epoch) + '.pth'))