-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdataset.py
115 lines (100 loc) · 4.79 KB
/
dataset.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
'''
**************************************************
@File ：AttitudeRecognition -> dataset
@IDE ：PyCharm
@Author ：TheOnlyMan
@Date ：2023/4/20 10:42
**************************************************
'''
import os.path
import random
from os.path import join
import cv2
import pandas as pd
import torch
from torch.utils.data import Dataset
from utils.image_utils import *
from utils.model_utils import switch
class DataSet(Dataset):
def __init__(self, args, flag='train', shape=(256, 192)) -> None:
self.args = args
self.flag = flag
self.shape = shape
self.out_shape = (shape[0] // 4, shape[1] // 4)
self.num_class = args.num_class
self.pixel_means = (np.array([102.9801, 115.9465, 122.7717], dtype=np.float32) / 255.0).reshape(3, 1, 1) # BGR
assert flag in ['train', 'valid', 'test', 'mytest'], 'not implement'
if self.flag == 'train':
self.path = os.path.join(args.dataset, 'train_image')
self.dtf = pd.read_csv(os.path.join(args.dataset, 'train_label.csv'))
else:
self.path = os.path.join(args.dataset, 'valid_image')
self.dtf = pd.read_csv(os.path.join(args.dataset, 'valid_label.csv'))
self.len = int(self.dtf.shape[0] * args.ratio / 100)
def flip(self, image, keypoints):
image = cv2.flip(image, 1)
keypoints[:, 0] = 1 - keypoints[:, 0]
keypoints = switch(keypoints)
return image, keypoints
def rotate(self, image, keypoints):
angle = random.uniform(0, 45)
if random.randint(0, 1):
angle *= -1
height, width = image.shape[:2]
center = (width / 2, height / 2)
rotMat = cv2.getRotationMatrix2D(center, angle, 1.0)
image = cv2.warpAffine(image, rotMat, (width, height))
for i in range(17):
x, y, v = keypoints[i]
coor = np.array([x * width, y * height, 1])
if v > 0:
coor = np.dot(rotMat, coor)
v *= ((coor[0] >= 0) & (coor[0] < width) & (coor[1] >= 0) & (coor[1] < height))
keypoints[i] = (coor[0] / width, coor[1] / height, v)
return image, keypoints
def __getitem__(self, item):
image_id = self.dtf.iloc[item, 0]
key_points = self.dtf.iloc[item, 1:18]
width, height, real_image_id, x_offset, y_offset = 0, 0, 0, 0, 0
if self.flag == 'test' or self.flag == 'mytest':
width, height, real_image_id = self.dtf.iloc[item, 18:21]
try:
x_offset, y_offset = self.dtf.iloc[item, 21:23]
except Exception: pass
key_points = np.array([np.array(i.strip('[]').split(', ')).astype(np.float32) for i in key_points])
image_path = join(self.path, f'{image_id}.jpg')
image = load_image(image_path)
new_image = cv2.resize(image, (192, 256), cv2.INTER_LANCZOS4)
if self.flag == 'train' and not self.args.noflip and random.randint(0, 3) == 0: # 0.25
new_image, key_points = self.flip(new_image, key_points)
elif self.flag == 'train' and not self.args.norotate and random.randint(0, 15) == 0: # 0.75 * 0.0625
new_image, key_points = self.rotate(new_image, key_points)
origin_keypoints = key_points.copy()
key_points[:, 0] = np.round(key_points[:, 0] * (self.shape[1] / 4)).astype(np.int32)
key_points[:, 1] = np.round(key_points[:, 1] * (self.shape[0] / 4)).astype(np.int32)
target15 = generate_label(key_points, self.num_class, (15, 15), self.out_shape)
target11 = generate_label(key_points, self.num_class, (11, 11), self.out_shape)
target9 = generate_label(key_points, self.num_class, (9, 9), self.out_shape)
target7 = generate_label(key_points, self.num_class, (7, 7), self.out_shape)
targets = [torch.Tensor(target15), torch.Tensor(target11), torch.Tensor(target9), torch.Tensor(target7)]
flip_image = new_image.copy()
input = im_to_torch(new_image)
input_flip = torch.Tensor([])
if self.flag != 'train' and not self.args.noflip:
flip_image = cv2.flip(flip_image, 1)
input_flip = im_to_torch(flip_image)
if self.flag == 'train':
input.mul_(random.uniform(0.8, 1.2)).clamp_(0, 1)
input -= self.pixel_means
if self.flag != 'train' and not self.args.noflip:
input_flip -= self.pixel_means
if self.flag == 'test':
return input, input_flip, width, height, real_image_id, x_offset, y_offset
elif self.flag == 'mytest':
return input, input_flip, width, height, image_id
elif self.flag == 'valid':
return input, input_flip, targets, origin_keypoints
elif self.flag == 'train':
return input, targets, origin_keypoints
def __len__(self):
return self.len