-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataloader.py
129 lines (110 loc) · 5.42 KB
/
dataloader.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
from pathlib import Path
from itertools import chain
import os
import random
from PIL import Image
import numpy as np
import torch
from torch.utils.data.sampler import WeightedRandomSampler
from torchvision import transforms
from munch import Munch
class InputFetcher:
def __init__(self, loader):
self.loader = loader
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
def _fetch_inputs(self):
try:
img, lbl, msk = next(self.iter)
except (AttributeError, StopIteration):
self.iter = iter(self.loader)
img, lbl, msk = next(self.iter)
return img, lbl, msk
def __next__(self):
img, lbl, msk = self._fetch_inputs()
inputs = Munch(image=img, label=lbl, mask=msk)
return Munch({k: v.to(self.device) for k, v in inputs.items()})
def listdir(dname):
#fnames = list(chain(*[list(Path(dname).rglob('*.' + ext)) for ext in ['jpg']]))
fnames = list(chain(*[list(Path(dname).rglob('*.' + ext)) for ext in ['png', 'jpg', 'jpeg', 'JPG']]))
return fnames
class dataset_loader(torch.utils.data.Dataset):
def __init__(self, args, dl_mode):
self.args = args
self.img_size = self.args.img_size
self.input_size = (self.img_size, self.img_size)
self.padding = 50
self.dl_mode = dl_mode
att_list = open(self.args.label_dir, 'r', encoding='utf-8').readlines()[1].split()
atts = [att_list.index(att) + 1 for att in self.args.attrs]
self.all_images_name = np.loadtxt(self.args.label_dir, skiprows=2, usecols=[0], dtype=np.str)
self.all_labels_attr = np.loadtxt(self.args.label_dir, skiprows=2, usecols=atts, dtype=np.int)
self.masks = self.load_masks(self.args.masks_dir)
if self.dl_mode == 'train':
self.image_dir = args.image_dir
self.images_name = self.all_images_name[:28000]
self.labels_attr = self.all_labels_attr[:28000]
elif self.dl_mode == 'val':
self.image_dir = args.image_val_dir
self.images_name = self.all_images_name[28000:]
self.labels_attr = self.all_labels_attr[28000:]
elif self.dl_mode == 'test':
att_list = open(self.args.test_label_dir, 'r', encoding='utf-8').readlines()[1].split()
atts = [att_list.index(att) + 1 for att in self.args.attrs]
self.images_name = np.loadtxt(self.args.test_label_dir, skiprows=2, usecols=[0], dtype=np.str)
self.labels_attr = np.loadtxt(self.args.test_label_dir, skiprows=2, usecols=atts, dtype=np.int)
self.image_dir = args.image_test_dir
self.masks = self.load_masks(self.args.masks_test_dir)
else:
print("Error")
self.length = len(self.images_name)
def load_masks(self, root):
fnames = os.listdir(root)
m_dir = []
for fname in sorted(fnames):
m_dir.append(os.path.join(root, fname))
return m_dir
def create_mask_box(self, width, height, mask_width, mask_height, x=None, y=None):
mask = np.ones((height, width))
mask_x = x if x is not None else random.randint(0 + self.padding, width - mask_width - self.padding)
mask_y = y if y is not None else random.randint(0 + self.padding, height - mask_height - self.padding)
mask[mask_y:mask_y + mask_height, mask_x:mask_x + mask_width] = 0
return mask
def __getitem__(self, index):
img = Image.open(os.path.join(self.image_dir, self.images_name[index]))
att = torch.tensor((self.labels_attr[index] + 1) // 2)
if self.dl_mode == 'test':
mask = Image.open(self.masks[index])
else:
ld_mask = Image.open(self.masks[index])
m_w, m_h = ld_mask.size
mask = ld_mask * self.create_mask_box(m_w, m_h, m_w // 3, m_h // 3)
mask = Image.fromarray(mask.astype('float32'))
if self.dl_mode == 'train':
self.img_transform = transforms.Compose([
transforms.Resize(size=self.input_size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5],
std=[0.5, 0.5, 0.5])
])
self.mask_transform = transforms.Compose([
transforms.Resize(size = self.input_size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor()
])
else:
self.img_transform = transforms.Compose([
transforms.Resize(size=self.input_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5],
std=[0.5, 0.5, 0.5])
])
self.mask_transform = transforms.Compose([
transforms.Resize(size=self.input_size),
transforms.ToTensor()
])
img = self.img_transform(img)
mask = self.mask_transform(mask)
return img, att, mask
def __len__(self):
return self.length