/
main.py
192 lines (160 loc) · 8.77 KB
/
main.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
import io
import os
from PIL import Image
import numpy as np
import time
import math
import matplotlib.pyplot as plt
import argparse
import torch
import torch.optim as optim
from torchvision import transforms
from torch.autograd import Variable
from deepvonet import DeepVONet
from dataset import VisualOdometryDataLoader
USE_CUDA = torch.cuda.is_available()
FLOAT = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
K = 100.
def to_tensor(ndarray, volatile=False, requires_grad=False, dtype=FLOAT):
return Variable(
torch.from_numpy(ndarray), volatile=volatile, requires_grad=requires_grad
).type(dtype)
def train_model(model, train_loader, criterion, optimizer, epoch, batch_size, trajectory_length):
# switch to train mode
for batch_idx, (images_stacked, odometries_stacked) in enumerate(train_loader):
if USE_CUDA:
images_stacked, odometries_stacked = images_stacked.cuda(), odometries_stacked.cuda()
images_stacked = images_stacked.permute(1, 0, 2, 3, 4)
images_stacked, odometries_stacked = Variable(images_stacked), Variable(odometries_stacked)
estimated_odometries = Variable(torch.zeros(odometries_stacked.shape))
estimated_odometries = estimated_odometries.permute(1, 0, 2)
if USE_CUDA:
estimated_odometries = estimated_odometries.cuda()
model.reset_hidden_states(size=batch_size, zero=True)
for t in range(trajectory_length):
# plt.imshow(images_stacked[0][0].data.cpu().numpy().transpose(1, 2, 0))
# plt.show()
# f, axarr = plt.subplots(2,2)
# axarr[0,0].imshow(image1[0].data.cpu().numpy().transpose((1, 2, 0)))
# axarr[0,1].imshow(image2[0].data.cpu().numpy().transpose((1, 2, 0)))
# plt.show()
# compute output
estimated_odometry = model(images_stacked[t])
estimated_odometries[t] = estimated_odometry
estimated_odometries = estimated_odometries.permute(1, 0, 2)
loss = criterion(estimated_odometries[:,:,:3], odometries_stacked[:,:,:3]) + K * criterion(estimated_odometries[:,:,3:], odometries_stacked[:,:,3:])
# compute gradient and do optimizer step
optimizer.zero_grad()
loss.backward()
optimizer.step()
print (epoch, batch_idx, loss.data.cpu()[0])
def train(model, datapath, checkpoint_path, epochs, trajectory_length, args):
model.train()
model.training = True
kwargs = {'num_workers': 1, 'pin_memory': True} if torch.cuda.is_available() else {}
train_loader = torch.utils.data.DataLoader(VisualOdometryDataLoader(datapath, trajectory_length=trajectory_length, transform=preprocess), batch_size=args.bsize, shuffle=True, drop_last=True, **kwargs)
criterion = torch.nn.MSELoss()
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.weight_decay)
for epoch in range(1, epochs + 1):
# train for one epoch
train_model(model, train_loader, criterion, optimizer, epoch, args.bsize, trajectory_length)
# # evaluate on validation set
# acc = test(test_loader, tripletnet, criterion, epoch)
#
# # remember best acc and save checkpoint
# is_best = acc > best_acc
# best_acc = max(acc, best_acc)
state = {
'epoch': epoch + 1,
'state_dict': model.state_dict(),
}
torch.save(state, os.path.join(checkpoint_path, "checkpoint_{}.pth".format(epoch)))
def test_model(model, test_loader, batch_size, trajectory_length):
for batch_idx, (images_stacked, odometries_stacked) in enumerate(test_loader):
if USE_CUDA:
images_stacked, odometries_stacked = images_stacked.cuda(), odometries_stacked.cuda()
images_stacked = images_stacked.permute(1, 0, 2, 3, 4)
images_stacked, odometries_stacked = Variable(images_stacked), Variable(odometries_stacked)
estimated_odometries = Variable(torch.zeros(odometries_stacked.shape))
estimated_odometries = estimated_odometries.permute(1, 0, 2)
if USE_CUDA:
estimated_odometries = estimated_odometries.cuda()
model.reset_hidden_states(size=batch_size, zero=True)
for t in range(trajectory_length):
estimated_odometry = model(images_stacked[t])
estimated_odometries[t] = estimated_odometry
estimated_odometries = estimated_odometries.permute(1, 0, 2)
print (estimated_odometries, odometries_stacked)
def test(model, datapath, trajectory_length, validation_steps, preprocess):
model.eval()
model.training = False
kwargs = {'num_workers': 1, 'pin_memory': True} if torch.cuda.is_available() else {}
test_loader = torch.utils.data.DataLoader(VisualOdometryDataLoader(datapath, trajectory_length=trajectory_length, transform=preprocess, test=True), batch_size=1, shuffle=True, **kwargs)
for epoch in range(1, validation_steps+1):
test_model(model, test_loader, 1, trajectory_length)
"""
with open(os.path.join(datapath, "index.txt"), 'r') as reader:
for index in reader:
index = index.strip()
images_path = []
with open(os.path.join(datapath, index, "index.txt"), 'r') as image_reader:
for image_path in image_reader:
images_path.append(image_path.strip())
model.reset_hidden_states(size=1, zero=True)
for image_index in range(len(images_path)-1):
model.reset_hidden_states(size=1, zero=False)
image1 = Image.open(os.path.join(datapath, index, images_path[image_index])).convert('RGB')
image2 = Image.open(os.path.join(datapath, index, images_path[image_index+1])).convert('RGB')
image1_tensor = preprocess(image1)
image2_tensor = preprocess(image2)
# plt.figure()
# plt.imshow(images_stacked.cpu().numpy().transpose((1, 2, 0)))
# plt.show()
images_stacked = torch.from_numpy(np.concatenate([image1_tensor, image2_tensor], axis=0))
images_stacked.unsqueeze_(0)
images_stacked = Variable(images_stacked).cuda()
odom = model(images_stacked)
print (image_index, image_index+1, odom.data.cpu())
del images_stacked, odom, image1_tensor, image2_tensor
"""
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='PyTorch on Place Recognition + Visual Odometry')
parser.add_argument('--mode', default='train', type=str, help='support option: train/test')
parser.add_argument('--datapath', default='datapath', type=str, help='path KITII odometry dataset')
parser.add_argument('--bsize', default=32, type=int, help='minibatch size')
parser.add_argument('--trajectory_length', default=10, type=int, help='trajectory length')
parser.add_argument('--lr', type=float, default=0.0001, metavar='LR', help='learning rate (default: 0.0001)')
parser.add_argument('--momentum', type=float, default=0.5, metavar='M', help='SGD momentum (default: 0.5)')
parser.add_argument('--weight_decay', type=float, default=1e-4, metavar='M', help='SGD momentum (default: 0.5)')
parser.add_argument('--tau', default=0.001, type=float, help='moving average for target network')
parser.add_argument('--debug', dest='debug', action='store_true')
parser.add_argument('--train_iter', default=20000000, type=int, help='train iters each timestep')
parser.add_argument('--validation_steps', default=100, type=int, help='test iters each timestep')
parser.add_argument('--epsilon', default=50000, type=int, help='linear decay of exploration policy')
parser.add_argument('--checkpoint_path', default=None, type=str, help='Checkpoint path')
parser.add_argument('--checkpoint', default=None, type=str, help='Checkpoint')
args = parser.parse_args()
normalize = transforms.Normalize(
#mean=[121.50361069 / 127., 122.37611083 / 127., 121.25987563 / 127.],
mean=[127. / 255., 127. / 255., 127. / 255.],
std=[1 / 255., 1 / 255., 1 / 255.]
)
preprocess = transforms.Compose([
transforms.Resize((384, 1280)),
transforms.CenterCrop((384, 1280)),
transforms.ToTensor(),
normalize
])
model = DeepVONet()
if args.checkpoint is not None:
checkpoint = torch.load(args.checkpoint)
model.load_state_dict(checkpoint['state_dict'])
if USE_CUDA:
model.cuda()
args = parser.parse_args()
if args.mode == 'train':
train(model, args.datapath, args.checkpoint_path, args.train_iter, args.trajectory_length, args)
elif args.mode == 'test':
test(model, args.datapath, args.trajectory_length, args.validation_steps, preprocess)
else:
raise RuntimeError('undefined mode {}'.format(args.mode))