-
Notifications
You must be signed in to change notification settings - Fork 6
/
create_dataset.py
162 lines (143 loc) · 6.85 KB
/
create_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
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
from torch.utils.data.dataset import Dataset
from PIL import Image
import torchvision
import torchvision.transforms.functional as TF
import os
import torch
import torch.nn.functional as F
import fnmatch
import numpy as np
import random
from utils import randrot,randfilp
from natsort import natsorted
class MSRSData(torch.utils.data.Dataset):
"""
Load dataset with infrared folder path and visible folder path
"""
# TODO: remove ground truth reference
def __init__(self, opts, is_train=True, crop=lambda x: x):
super(MSRSData, self).__init__()
self.is_train = is_train
if is_train:
self.vis_folder = os.path.join(opts.dataroot, 'train', 'vi')
self.ir_folder = os.path.join(opts.dataroot, 'train', 'ir')
self.label_folder = os.path.join(opts.dataroot, 'train', 'label')
self.bi_folder = os.path.join(opts.dataroot, 'train', 'bi')
self.bd_folder = os.path.join(opts.dataroot, 'train', 'bd')
self.mask_folder = os.path.join(opts.dataroot, 'train', 'mask')
else:
self.vis_folder = os.path.join(opts.dataroot, 'test', 'vi')
self.ir_folder = os.path.join(opts.dataroot, 'test', 'ir')
self.label_folder = os.path.join(opts.dataroot, 'test', 'label')
self.crop = torchvision.transforms.RandomCrop(256)
# gain infrared and visible images list
self.ir_list = natsorted(os.listdir(self.label_folder))
print(len(self.ir_list))
#self.ST = SpatialTransformer(self.crop.size[0],self.crop.size[0],False)
def __getitem__(self, index):
# gain image path
image_name = self.ir_list[index]
vis_path = os.path.join(self.vis_folder, image_name)
ir_path = os.path.join(self.ir_folder, image_name)
label_path = os.path.join(self.label_folder, image_name)
# read image as type Tensor
vis = self.imread(path=vis_path)
ir = self.imread(path=ir_path, vis_flage=False)
label = self.imread(path=label_path, label=True)
if self.is_train:
bi_path = os.path.join(self.bi_folder, image_name)
bd_path = os.path.join(self.bd_folder, image_name)
mask_path = os.path.join(self.mask_folder, image_name)
bi = self.imread(path=bi_path, label=True)
bd = self.imread(path=bd_path, label=True)
mask = self.imread(path=mask_path, vis_flage=False)
if self.is_train:
## 训练图像进行一定的数据增强，包括翻转，旋转，以及随机裁剪等
vis_ir = torch.cat([vis, ir, label, bi, bd, mask],dim=1)
if vis_ir.shape[-1]<=256 or vis_ir.shape[-2]<=256:
vis_ir=TF.resize(vis_ir,256)
vis_ir = randfilp(vis_ir)
vis_ir = randrot(vis_ir)
patch = self.crop(vis_ir)
vis, ir, label, bi, bd, mask = torch.split(patch, [3, 1, 1, 1, 1, 1], dim=1)
label = label.type(torch.LongTensor)
bi = bi / 255.0
bd = bd / 255.0
bi = bi.type(torch.LongTensor)
bd = bd.type(torch.LongTensor)
## ir 单通道, vis RGB三通道
return ir.squeeze(0), vis.squeeze(0), label.squeeze(0), bi.squeeze(0), bd.squeeze(0), mask.squeeze(0)
else:
label = label.type(torch.LongTensor)
return ir.squeeze(0), vis.squeeze(0), label.squeeze(0), image_name
def __len__(self):
return len(self.ir_list)
@staticmethod
def imread(path, label=False, vis_flage=True):
if label:
img = Image.open(path)
im_ts = TF.to_tensor(img).unsqueeze(0) * 255
else:
if vis_flage: ## visible images; RGB channel
img = Image.open(path).convert('RGB')
im_ts = TF.to_tensor(img).unsqueeze(0)
else: ## infrared images single channel
img = Image.open(path).convert('L')
im_ts = TF.to_tensor(img).unsqueeze(0)
return im_ts
class FusionData(torch.utils.data.Dataset):
"""
Load dataset with infrared folder path and visible folder path
"""
def __init__(self, opts, crop=lambda x: x):
super(FusionData, self).__init__()
self.vis_folder = os.path.join(opts.dataroot, opts.dataname, 'test', 'vi')
self.ir_folder = os.path.join(opts.dataroot, opts.dataname, 'test', 'ir')
# self.vis_folder = os.path.join(opts.dataroot, opts.dataname, 'vi')
# self.ir_folder = os.path.join(opts.dataroot, opts.dataname, 'ir')
# self.vis_folder = os.path.join('/data/timer/Segmentation/SegNext/datasets/MSRS/RGB')
# self.ir_folder = os.path.join('/data/timer/Segmentation/SegNext/datasets/MSRS/Thermal')
self.ir_list = natsorted(os.listdir(self.ir_folder))
print(len(self.ir_list))
def __getitem__(self, index):
# gain image path
image_name = self.ir_list[index]
vis_path = os.path.join(self.vis_folder, image_name)
ir_path = os.path.join(self.ir_folder, image_name)
# read image as type Tensor
vis, w, h = self.imread(path=vis_path)
ir, w, h = self.imread(path=ir_path, vis_flage=False)
return ir.squeeze(0), vis.squeeze(0), image_name, w, h
def __len__(self):
return len(self.ir_list)
@staticmethod
def imread(path, label=False, vis_flage=True):
if label:
img = Image.open(path)
# 获取图像大小
width, height = img.size
# 调整图像大小到32的倍数
new_width = width - (width % 32)
new_height = height - (height % 32)
img = img.resize((new_width, new_height))
im_ts = TF.to_tensor(img).unsqueeze(0) * 255
else:
if vis_flage: ## visible images; RGB channel
img = Image.open(path).convert('RGB')
# 获取图像大小
width, height = img.size
# 调整图像大小到32的倍数
new_width = width - (width % 32)
new_height = height - (height % 32)
img = img.resize((new_width, new_height))
im_ts = TF.to_tensor(img).unsqueeze(0)
else: ## infrared images single channel
img = Image.open(path).convert('L')
# 获取图像大小
width, height = img.size
# 调整图像大小到32的倍数
new_width = width - (width % 32)
new_height = height - (height % 32)
img = img.resize((new_width, new_height))
im_ts = TF.to_tensor(img).unsqueeze(0)
return im_ts, width, height