-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocess.py
129 lines (111 loc) · 4.59 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
120
121
122
123
124
125
126
127
128
129
import os
import numpy as np
import pandas as pd
import cv2
from tqdm import trange, tqdm
import config
import argparse
def gaussian(x, mu, sig):
return (
np.exp(-np.power((x - mu) / (sig / 2), 2.0) / 2)
)
def convert_y(input):
meshgrid, _, _ = np.meshgrid(np.linspace(0, config.classes, config.classes + 1), np.linspace(0, 0, 640), np.linspace(0, 0, 360), indexing='ij')
output = np.empty((config.classes + 1, 640, 360))
for i in range(config.classes + 1):
output[i] = (meshgrid[i] == input)
return output
def get_y(visibility, x_coord, y_coord, std):
x, y = np.indices([640, 360])
if (visibility == 1 or visibility == 2 or visibility == 3):
x_coord = round(x_coord / 2)
y_coord = round(y_coord / 2)
x = gaussian(x, x_coord, std)
y = gaussian(y, y_coord, std)
if (visibility == 0):
x = np.zeros((640, 360))
y = np.zeros((640, 360))
# print(visibility)
output = x * y
# output *= (2 * np.pi * (std)**2) # seems to work if std is squared but not squared in paper (squared because gaussian generation is different?)
output *= config.classes
# print(output.shape, output.max())
output = np.rint(output).astype(np.int16)
return output
def get_x(imgs):
# print(imgs[0].shape)
# print(imgs[1].shape)
# print(imgs[2].shape)
a = cv2.resize(imgs[0], (640, 360)).swapaxes(0, 2)
b = cv2.resize(imgs[1], (640, 360)).swapaxes(0, 2)
c = cv2.resize(imgs[2], (640, 360)).swapaxes(0, 2)
output = np.vstack((a, b, c)).astype(np.float32)
return output
def extract_data(csv_path, video_path):
x = []
y = []
df = pd.read_csv(csv_path).to_numpy()
cap = cv2.VideoCapture(video_path)
length = len(df)
q = [np.zeros((720, 1280, 3)), np.zeros((720, 1280, 3)), np.zeros((720, 1280, 3))]
for i in range(length):
_, frame = cap.read()
q.pop(0)
q.append(frame)
if (i % 10 == 0 and df[i - 1][1] == 1):
x.append(get_x(q))
y.append(get_y(df[i - 1][1], df[i - 1][2], df[i - 1][3], std=5))
x = np.array(x, dtype=np.float32)
y = np.array(y, dtype=np.int16)
return x, y
def Badmintonpreprocess(data_path, write_path):
csv_paths = []
video_paths = []
for match in os.listdir(data_path):
for path in os.listdir(os.path.join(data_path, match, 'csv')):
csv_paths.append(os.path.join(data_path, match, 'csv', path))
video_paths.append(os.path.join(data_path, match, 'video', path.replace('_ball.csv', '.mp4')))
index = 0
for i in trange(len(csv_paths)):
csv_path = csv_paths[i]
video_path = video_paths[i]
x_imgs, y_imgs = extract_data(csv_path, video_path)
for x_img, y_img in zip(x_imgs, y_imgs):
np.save(os.path.join(write_path, 'imgs', str(index)) + '.npy', x_img)
np.save(os.path.join(write_path, 'labels', str(index)) + '.npy', y_img)
index += 1
def Tennispreprocess(data_path, write_path):
index = 0
for game in tqdm(os.listdir(data_path)):
for clip in os.listdir(os.path.join(data_path, game)):
q = []
df = pd.read_csv(os.path.join(data_path, game, clip, 'Label.csv'))
for i in range(len(os.listdir(os.path.join(data_path, game, clip))) - 1):
q.append(cv2.imread(os.path.join(data_path, game, clip, f"{i:04d}.jpg")))
if (len(q) == 3):
np.save(os.path.join(write_path, 'imgs', f"{str(index)}.npy"), get_x(q))
np.save(os.path.join(write_path, 'labels', f"{str(index)}.npy"), get_y(df.iloc[i]['visibility'], df.iloc[i]['x-coordinate'], df.iloc[i]['y-coordinate'], 5))
q.pop(0)
index += 1
return
def clear(path):
filelist = [ f for f in os.listdir(path) if f.endswith(".npy") ]
for f in filelist:
os.remove(os.path.join(path, f))
def main():
parser = argparse.ArgumentParser()
parser.add_argument('type', help='b=Badminton Dataset; t=Tennis Dataset')
args = parser.parse_args()
assert args.type == 'b' or args.type == 't', 'Wrong arguments'
clear('data/train/imgs')
clear('data/train/labels')
clear('data/valid/imgs')
clear('data/valid/labels')
if (args.type == 'b'):
Badmintonpreprocess('raw_data/train', 'data/train')
Badmintonpreprocess('raw_data/valid', 'data/valid')
if (args.type == 't'):
Tennispreprocess('raw_data_tennis/train', 'data/train')
Tennispreprocess('raw_data_tennis/valid', 'data/valid')
if __name__=='__main__':
main()