-
Notifications
You must be signed in to change notification settings - Fork 34
/
nyuDataLoader.py
173 lines (136 loc) · 6.13 KB
/
nyuDataLoader.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
163
164
165
166
167
168
169
170
171
172
173
import glob
import numpy as np
import os.path as osp
from PIL import Image
import random
import struct
from torch.utils.data import Dataset
import scipy.ndimage as ndimage
import cv2
from skimage.measure import block_reduce
import json
import scipy.ndimage as ndimage
class ConcatDataset(Dataset ):
def __init__(self, *datasets):
self.datasets = datasets
def __getitem__(self, i):
return tuple(d[i] for d in self.datasets )
def __len__(self ):
return max(len(d) for d in self.datasets )
class NYULoader(Dataset ):
def __init__(self, imRoot, normalRoot, depthRoot, segRoot,
imHeight = 480, imWidth = 640,
imWidthMax = 600, imWidthMin = 560,
phase='TRAIN', rseed = None ):
self.imRoot = imRoot
self.imHeight = imHeight
self.imWidth = imWidth
self.phase = phase.upper()
self.imWidthMax = imWidthMax
self.imWidthMin = imWidthMin
if phase == 'TRAIN':
with open('NYUTrain.txt', 'r') as fIn:
imList = fIn.readlines()
self.imList = [osp.join(self.imRoot, x.strip() ) for x in imList ]
elif phase == 'TEST':
with open('NYUTest.txt', 'r') as fIn:
imList = fIn.readlines()
self.imList = [osp.join(self.imRoot, x.strip() ) for x in imList ]
self.normalList = [x.replace(imRoot, normalRoot) for x in self.imList ]
self.segList = [x.replace(imRoot, segRoot) for x in self.imList ]
self.depthList = [x.replace(imRoot, depthRoot).replace('.png', '.tiff') for x in self.imList]
print('Image Num: %d' % len(self.imList) )
# Permute the image list
self.count = len(self.imList )
self.perm = list(range(self.count ) )
if rseed is not None:
random.seed(0)
random.shuffle(self.perm )
def __len__(self):
return len(self.perm )
def __getitem__(self, ind):
ind = (ind % len(self.perm) )
if ind == 0:
random.shuffle(self.perm )
if self.phase == 'TRAIN':
scale = np.random.random();
imCropWidth = int( np.round( (self.imWidthMax - self.imWidthMin ) * scale + self.imWidthMin ) )
imCropHeight = int( float(self.imHeight) / float(self.imWidth ) * imCropWidth )
rs = int(np.round( (480 - imCropHeight) * np.random.random() ) )
re = rs + imCropHeight
cs = int(np.round( (640 - imCropWidth) * np.random.random() ) )
ce = cs + imCropWidth
elif self.phase == 'TEST':
imCropWidth = self.imWidth
imCropHeight = self.imHeight
rs, re, cs, ce = 0, 480, 0, 640
segNormal = 0.5 * ( self.loadImage(self.segList[self.perm[ind] ], rs, re, cs, ce) + 1)[0:1, :, :]
# Read Image
im = 0.5 * (self.loadImage(self.imList[self.perm[ind] ], rs, re, cs, ce, isGama = True ) + 1)
# normalize the normal vector so that it will be unit length
normal = self.loadImage( self.normalList[self.perm[ind] ], rs, re, cs, ce )
normal = normal / np.sqrt(np.maximum(np.sum(normal * normal, axis=0), 1e-5) )[np.newaxis, :]
# Read depth
depth = self.loadDepth(self.depthList[self.perm[ind] ], rs, re, cs, ce )
if imCropHeight != self.imHeight or imCropWidth != self.imWidth:
depth = np.squeeze(depth, axis=0)
depth = cv2.resize(depth, (self.imWidth, self.imHeight), interpolation = cv2.INTER_LINEAR)
depth = depth[np.newaxis, :, :]
segDepth = np.logical_and(depth > 1, depth < 10).astype(np.float32 )
if imCropHeight != self.imHeight or imCropWidth != self.imWidth:
normal = normal.transpose([1, 2, 0] )
normal = cv2.resize(normal, (self.imWidth, self.imHeight), interpolation = cv2.INTER_LINEAR)
normal = normal.transpose([2, 0, 1] )
normal = normal / np.maximum(np.sqrt(np.sum(normal * normal, axis=0 )[np.newaxis, :, :] ), 1e-5)
if imCropHeight != self.imHeight or imCropWidth != self.imWidth:
segNormal = np.squeeze(segNormal, axis=0)
segNormal = cv2.resize(segNormal, (self.imWidth, self.imHeight), interpolation = cv2.INTER_LINEAR)
segNormal = segNormal[np.newaxis, :, :]
im = im.transpose([1, 2, 0] )
im = cv2.resize(im, (self.imWidth, self.imHeight), interpolation = cv2.INTER_LINEAR )
im = im.transpose([2, 0, 1] )
if self.phase == 'TRAIN':
if np.random.random() > 0.5:
normal = np.ascontiguousarray(normal[:, :, ::-1] )
normal[0, :, :] = -normal[0, :, :]
depth = np.ascontiguousarray(depth[:, :, ::-1] )
segNormal = np.ascontiguousarray(segNormal[:, :, ::-1] )
segDepth = np.ascontiguousarray(segDepth[:, :, ::-1] )
im = np.ascontiguousarray(im[:, :, ::-1] )
scale = 1 + ( np.random.random(3) * 0.4 - 0.2 )
scale = scale.reshape([3, 1, 1] )
im = im * scale
batchDict = {'normal': normal,
'depth': depth,
'segNormal': segNormal,
'segDepth': segDepth,
'im': im.astype(np.float32 ),
'name': self.imList[self.perm[ind] ]
}
return batchDict
def loadImage(self, imName, rs, re, cs, ce, isGama = False):
if not(osp.isfile(imName ) ):
print(imName )
assert(False )
im = cv2.imread(imName)
if len(im.shape ) == 3:
im = im[:, :, ::-1]
im = im[rs:re, cs:ce, :]
im = np.ascontiguousarray(im.astype(np.float32 ) )
if isGama:
im = (im / 255.0) ** 2.2
im = 2 * im - 1
else:
im = (im - 127.5) / 127.5
if len(im.shape) == 2:
im = im[:, np.newaxis]
im = np.transpose(im, [2, 0, 1] )
return im
def loadDepth(self, imName, rs, re, cs, ce ):
if not osp.isfile(imName):
print(imName )
assert(False )
im = cv2.imread(imName, -1)
im = im[rs:re, cs:ce]
im = im[np.newaxis, :, :]
return im