-
Notifications
You must be signed in to change notification settings - Fork 7
/
train_lanenet-Copy.py
165 lines (147 loc) · 7.99 KB
/
train_lanenet-Copy.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
import os
import os.path as ops
import argparse
import time
import math
import sys
#print(sys.path)
import glob
import numpy as np
import cv2
import torch
import torch.nn.functional as F
from lanenet_model.lanenet_merge_model import *
from lanenet_model.loss import *
from utils.data_loader import LaneDataSet
from torch.utils.data import DataLoader
from torch.autograd import Variable
from lanenet_model.discriminative_loss import *
#setting state to adjust learning rate
state = {}
state["schedule"] = [500,1000]
state["lr"] = 0.01
def adjust_learning_rate(optimizer, epoch):
global state
if epoch in state["schedule"]:
state["lr"] *= 0.1
for param_group in optimizer.param_groups:
param_group['lr'] = state['lr']
def save_model(path,epoch,model):
if isinstance(model,nn.DataParallel):
save = model.module
else:
save = model
torch.save(save.state_dict(),os.path.join(path,"train_{}_epochs.pkl".format(epoch+1)))
def train_net(data_dir, save_path,im_path,train_epochs,pretrained_model = None,net_flag='vgg',use_gpu = True,gpu_count = 8):
#Todo: adopt from get_train_val_loader
train_dataset = LaneDataSet(data_dir,"train")
val_dataset = LaneDataSet(data_dir,"val")
model = LaneNet(net_flag)
start_epoch = 0
if pretrained_model:
saved = torch.load(pretrained_model)
model.load_state_dict({k:v for k,v in saved.items() if k in model.state_dict()})
#start_epoch = int(pretrained_model.split("_")[-2])
if use_gpu:
model = nn.DataParallel(model).cuda()
optimizer = torch.optim.SGD(model.parameters(),lr = state['lr'],momentum=0.9,weight_decay=5e-4)
# optimizer = torch.optim.Adam(model.parameters(),lr = 0.0005)
if not os.path.exists(save_path):
os.makedirs(save_path)
if not os.path.exists(im_path):
os.makedirs(im_path)
#print("Global configuration is as follows:")
global_step = 0
k_binary_loss = 1.
k_instance_loss = 0.
for epoch in range(start_epoch,train_epochs):
t_start = time.time()
train_loader = DataLoader(train_dataset,batch_size=4,pin_memory=True)
adjust_learning_rate(optimizer,epoch)
phase = "train"
model.train()
for input_data in train_loader:
optimizer.zero_grad()
global_step += 1
image_data = Variable(input_data["input_tensor"]).cuda().type(torch.cuda.FloatTensor)
instance_label = Variable(input_data["instance_label"]).cuda().type(torch.cuda.LongTensor)
binary_label = Variable(input_data["binary_label"]).cuda().type(torch.cuda.LongTensor)
#output process
net_output = model(image_data)
binary_seg_logits = net_output["binary_seg_logits"]
#binary_seg_logits = torch.sigmoid(binary_seg_logits)
#binary_loss = dice_loss_single(binary_seg_logits,binary_label)
loss_fn = nn.CrossEntropyLoss()
binary_loss = loss_fn(binary_seg_logits,binary_label)
pix_embedding = net_output["instance_seg_logits"]
instance_loss, _, _, _ = discriminative_loss(pix_embedding,instance_label, 3, 0.5, 1.5, 1.0, 1.0, 0.001)
total_loss = binary_loss * k_binary_loss + instance_loss * k_instance_loss
out = net_output["binary_seg_pred"]
binary_label = binary_label.unsqueeze(1).type(torch.cuda.ByteTensor)
pix_cls = out.masked_select(binary_label)
train_acc = pix_cls.nonzero().size()[0]/pix_cls.size()[0]
if np.isnan(total_loss.item()) or np.isnan(binary_loss.item()) or np.isnan(instance_loss.item()):
print('cost is: {:.5f}'.format(total_loss.item()))
print('binary cost is: {:.5f}'.format(binary_loss.item()))
print('instance cost is: {:.5f}'.format(instance_loss.item()))
cv2.imwrite('nan_image.png', image_data[0].cpu().numpy().transpose(1,2,0) + VGG_MEAN)
cv2.imwrite('nan_instance_label.png', image_data[0].cpu().numpy().transpose(1,2,0))
cv2.imwrite('nan_binary_label.png', binary_label[0].cpu().numpy().transpose(1,2,0) * 255)
cv2.imwrite('nan_embedding.png', pix_embedding[0].cpu().numpy().transpose(1,2,0))
break
print("Epoch: {}, Elapsed Time: {:.2f}".format(epoch+1, (time.time() - t_start)/60.0),"total loss {:.5f}".format(total_loss.item()),"binary loss {:.5f}".format(binary_loss.item() * k_binary_loss),"instance loss {:.5f}".format(instance_loss.item() * k_instance_loss),"train accuracy {:.5f}".format(train_acc))
cv2.imwrite(os.path.join(im_path,str(epoch+1)+"train_pred.png"),out[0].squeeze(0).cpu().numpy().transpose(0,1)*255)
cv2.imwrite(os.path.join(im_path,str(epoch+1)+"train_label.png"),binary_label[0].squeeze(0).cpu().numpy().transpose(0,1)*255)
total_loss.backward()
optimizer.step()
if (epoch+1)%10 == 0:
save_model(save_path,epoch,model)
#valid phase
phase = "valid"
valid_loader = DataLoader(val_dataset,batch_size=4,pin_memory=True)
for input_data in valid_loader:
image_data = Variable(input_data["input_tensor"]).cuda().type(torch.cuda.FloatTensor)
instance_label = Variable(input_data["instance_label"]).cuda().type(torch.cuda.LongTensor)
binary_label = Variable(input_data["binary_label"]).type(torch.cuda.LongTensor)
#output process
val_acc = []
val_total_loss_list = []
val_binary_loss_list = []
val_instance_loss_list = []
net_output = model(image_data)
binary_seg_logits = net_output["binary_seg_logits"]
#binary_seg_logits = torch.sigmoid(binary_seg_logits)
#val_binary_loss = dice_loss_single(binary_seg_logits, binary_label)
loss_fn = nn.CrossEntropyLoss()
val_binary_loss = loss_fn(binary_seg_logits,binary_label)
pix_embedding = net_output["instance_seg_logits"]
val_instance_loss, _, _, _ = discriminative_loss(pix_embedding, instance_label, 3, 0.5, 1.5, 1.0, 1.0, 0.001)
val_total_loss = val_binary_loss*1.0 + val_instance_loss * 0.1
val_total_loss_list.append(val_total_loss.item())
val_binary_loss_list.append(val_binary_loss.item())
val_instance_loss_list.append(val_instance_loss.item())
out = net_output["binary_seg_pred"]
binary_label = binary_label.unsqueeze(1).type(torch.cuda.ByteTensor)
pix_cls = out.masked_select(binary_label)
val_acc.append(pix_cls.nonzero().size()[0]/pix_cls.size()[0])
print("Epoch {} Validation Report".format(epoch+1),"total loss {:.5f}".format(np.mean(val_total_loss_list)),"binary loss {:.5f}".format(np.mean(val_binary_loss_list)),"instance loss {:.5f}".format(np.mean(val_instance_loss_list)),"valid accuracy {:.5f}".format(np.mean(val_acc)))
# if ((epoch+1)%100 == 0):
cv2.imwrite(os.path.join(im_path,str(epoch+1)+'binary_pred_label.png'), out[0].squeeze(0).cpu().numpy().transpose(0,1) * 255)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("command",help="train or test on lanenet model")
parser.add_argument("--dataset",help="Directory of the AIC dataset")
parser.add_argument("--save",help="Directory to save model checkpoint")
parser.add_argument("--epochs",type=int,help="Training epochs")
parser.add_argument("--pretrained",required = False,default=None,help="pretrained model path")
parser.add_argument("--net",help = "backbone net")
parser.add_argument("--image",help = "output val image")
# parser.add_argument("--batch_size",help="Training or testing batch size")
# parser.add_argument("--gpu",help="use gpu or not")
return parser.parse_args()
if __name__ == '__main__':
args = parse_args()
command = args.command
save_path = args.save
if command == 'train':
train_net(args.dataset,save_path,args.image,args.epochs,args.pretrained,net_flag = args.net)