-
Notifications
You must be signed in to change notification settings - Fork 107
/
preprocess.py
120 lines (108 loc) · 3.99 KB
/
preprocess.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
import os
import glob
import numpy as np
import time
from helper import R_to_angle
from params import par
from torchvision import transforms
from PIL import Image
import torch
import math
def clean_unused_images():
seq_frame = {'00': ['000', '004540'],
'01': ['000', '001100'],
'02': ['000', '004660'],
'03': ['000', '000800'],
'04': ['000', '000270'],
'05': ['000', '002760'],
'06': ['000', '001100'],
'07': ['000', '001100'],
'08': ['001100', '005170'],
'09': ['000', '001590'],
'10': ['000', '001200']
}
for dir_id, img_ids in seq_frame.items():
dir_path = '{}{}/'.format(par.image_dir, dir_id)
if not os.path.exists(dir_path):
continue
print('Cleaning {} directory'.format(dir_id))
start, end = img_ids
start, end = int(start), int(end)
for idx in range(0, start):
img_name = '{:010d}.png'.format(idx)
img_path = '{}{}/{}'.format(par.image_dir, dir_id, img_name)
if os.path.isfile(img_path):
os.remove(img_path)
for idx in range(end+1, 10000):
img_name = '{:010d}.png'.format(idx)
img_path = '{}{}/{}'.format(par.image_dir, dir_id, img_name)
if os.path.isfile(img_path):
os.remove(img_path)
# transform poseGT [R|t] to [theta_x, theta_y, theta_z, x, y, z]
# save as .npy file
def create_pose_data():
info = {'00': [0, 4540], '01': [0, 1100], '02': [0, 4660], '03': [0, 800], '04': [0, 270], '05': [0, 2760], '06': [0, 1100], '07': [0, 1100], '08': [1100, 5170], '09': [0, 1590], '10': [0, 1200]}
start_t = time.time()
for video in info.keys():
fn = '{}{}.txt'.format(par.pose_dir, video)
print('Transforming {}...'.format(fn))
with open(fn) as f:
lines = [line.split('\n')[0] for line in f.readlines()]
poses = [ R_to_angle([float(value) for value in l.split(' ')]) for l in lines] # list of pose (pose=list of 12 floats)
poses = np.array(poses)
base_fn = os.path.splitext(fn)[0]
np.save(base_fn+'.npy', poses)
print('Video {}: shape={}'.format(video, poses.shape))
print('elapsed time = {}'.format(time.time()-start_t))
def calculate_rgb_mean_std(image_path_list, minus_point_5=False):
n_images = len(image_path_list)
cnt_pixels = 0
print('Numbers of frames in training dataset: {}'.format(n_images))
mean_np = [0, 0, 0]
mean_tensor = [0, 0, 0]
to_tensor = transforms.ToTensor()
image_sequence = []
for idx, img_path in enumerate(image_path_list):
print('{} / {}'.format(idx, n_images), end='\r')
img_as_img = Image.open(img_path)
img_as_tensor = to_tensor(img_as_img)
if minus_point_5:
img_as_tensor = img_as_tensor - 0.5
img_as_np = np.array(img_as_img)
img_as_np = np.rollaxis(img_as_np, 2, 0)
cnt_pixels += img_as_np.shape[1]*img_as_np.shape[2]
for c in range(3):
mean_tensor[c] += float(torch.sum(img_as_tensor[c]))
mean_np[c] += float(np.sum(img_as_np[c]))
mean_tensor = [v / cnt_pixels for v in mean_tensor]
mean_np = [v / cnt_pixels for v in mean_np]
print('mean_tensor = ', mean_tensor)
print('mean_np = ', mean_np)
std_tensor = [0, 0, 0]
std_np = [0, 0, 0]
for idx, img_path in enumerate(image_path_list):
print('{} / {}'.format(idx, n_images), end='\r')
img_as_img = Image.open(img_path)
img_as_tensor = to_tensor(img_as_img)
if minus_point_5:
img_as_tensor = img_as_tensor - 0.5
img_as_np = np.array(img_as_img)
img_as_np = np.rollaxis(img_as_np, 2, 0)
for c in range(3):
tmp = (img_as_tensor[c] - mean_tensor[c])**2
std_tensor[c] += float(torch.sum(tmp))
tmp = (img_as_np[c] - mean_np[c])**2
std_np[c] += float(np.sum(tmp))
std_tensor = [math.sqrt(v / cnt_pixels) for v in std_tensor]
std_np = [math.sqrt(v / cnt_pixels) for v in std_np]
print('std_tensor = ', std_tensor)
print('std_np = ', std_np)
if __name__ == '__main__':
clean_unused_images()
create_pose_data()
# Calculate RGB means of images in training videos
train_video = ['00', '02', '08', '09', '06', '04', '10']
image_path_list = []
for folder in train_video:
image_path_list += glob.glob('KITTI/images/{}/*.png'.format(folder))
calculate_rgb_mean_std(image_path_list, minus_point_5=True)