-
Notifications
You must be signed in to change notification settings - Fork 13
/
imagenet.py
228 lines (177 loc) · 7.94 KB
/
imagenet.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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
"""
File: imagenet.py
Author: Nrupatunga
Email: nrupatunga.s@byjus.com
Github: https://github.com/nrupatunga
Description: loading imagenet dataset
"""
import itertools
import sys
import xml.etree.ElementTree as ET
from pathlib import Path
import numpy as np
from loguru import logger
try:
from goturn.helper import image_io
from goturn.helper.annotation import annotation
from goturn.helper.vis_utils import Visualizer
from goturn.dataloaders.sampler import sample_generator
except ImportError:
logger.error('Please run $source settings.sh from root directory')
sys.exit(1)
class ImageNetDataset:
def __init__(self, imgs_dir, ann_dir, isTrain=True, val_ratio=0.2, dbg=False):
'''
loading images and annotation from imagenet
@imgs_dir: images path
@ann_dir: annotations path
@isTrain: True: Training, False: validation
@val_ratio: validation data ratio
@dbg: For visualization
'''
if not Path(imgs_dir).is_dir():
logger.error('{} is not a valid directory'.format(imgs_dir))
self._imgs_dir = Path(imgs_dir)
self._ann_dir = Path(ann_dir)
self._kMaxRatio = 0.66
self._list_of_annotations = self.__loadImageNetDet(isTrain=isTrain, val_ratio=val_ratio)
self._data_fetched = [] # for debug purposes
assert len(self._list_of_annotations) > 0, 'Number of valid annotations is {}'.format(len(self._list_of_annotations))
self._dbg = dbg
if dbg:
self._env = 'ImageNet'
self._viz = Visualizer(env=self._env)
def __len__(self):
''' number of valid annotations that is fetched in
load_annotation_file'''
return len(self._list_of_annotations)
def __getitem__(self, idx):
"""Get the current idx data
@idx: Current index for the data
"""
# set the dbg to true, to see if all the images are fetched
# atleast once during the entire epoch
image, bbox = self.__load_annotation(idx)
# For imagenet, prev frame and current frame are the same
return image, bbox, image, bbox
def __loadImageNetDet(self, isTrain=True, val_ratio=0.2):
''' Loads all the image annotations files '''
subdirs = [x.parts[-1] for x in self._imgs_dir.iterdir() if x.is_dir()]
train_ratio = 1 - val_ratio
num_end = int(train_ratio * len(subdirs))
if isTrain:
subdirs = subdirs[0:num_end]
else:
subdirs = subdirs[num_end:]
self._subdirs = subdirs
num_annotations = 0
list_of_annotations_out = []
for i, subdir in enumerate(subdirs):
ann_files = self._ann_dir.joinpath(subdir).glob('*.xml')
logger.info('Loading {}/{} - annotation file from folder = {}'.format(i + 1, len(subdirs), subdir))
for ann in ann_files:
list_of_annotations, num_ann_curr = self.__load_annotation_file(ann)
num_annotations = num_annotations + num_ann_curr
if len(list_of_annotations) == 0:
continue
list_of_annotations_out.append(list_of_annotations)
all_annotations = list(itertools.chain.from_iterable(list_of_annotations_out))
# random.shuffle(all_annotations)
logger.info('+' * 60)
logger.info("Found {} annotations from {} images"
" ({:.2f} annotations/image)".format(num_annotations,
len(list_of_annotations_out),
(num_annotations / len(list_of_annotations_out))))
return all_annotations
def __load_annotation_file(self, annotation_file):
""" Loads the bounding box annotations in xml file
@annotation_file: annotation file (.xml), which contains
bounding box information
"""
list_of_annotations = []
num_annotations = 0
root = ET.parse(annotation_file).getroot()
folder = root.find('folder').text
filename = root.find('filename').text
size = root.find('size')
disp_width = int(size.find('width').text)
disp_height = int(size.find('height').text)
for obj in root.findall('object'):
bbox = obj.find('bndbox')
xmin = int(bbox.find('xmin').text)
xmax = int(bbox.find('xmax').text)
ymin = int(bbox.find('ymin').text)
ymax = int(bbox.find('ymax').text)
width = xmax - xmin
height = ymax - ymin
kMaxRatio = self._kMaxRatio
if width > (kMaxRatio * disp_width) or height > (kMaxRatio * disp_height):
continue
if ((xmin < 0) or (ymin < 0) or (xmax <= xmin) or (ymax <= ymin)):
continue
objAnnotation = annotation()
objAnnotation.setbbox(xmin, xmax, ymin, ymax)
objAnnotation.setWidthHeight(disp_width, disp_height)
objAnnotation.setImagePath(Path(folder).joinpath(filename))
list_of_annotations.append(objAnnotation)
num_annotations = num_annotations + 1
return list_of_annotations, num_annotations
def __load_annotation(self, idx):
"""
this loads the image and its corresponding bounding box gt
@idx: current image number to be fetched
"""
random_ann = self._list_of_annotations[idx]
img_path = self._imgs_dir.joinpath(random_ann.image_path.with_suffix('.JPEG'))
image = image_io.load(img_path)
image = np.asarray(image, dtype=np.uint8)
img_height = image.shape[0]
img_width = image.shape[1]
sc_factor_1 = 1.0
if img_height != random_ann.disp_height or img_width != random_ann.disp_width:
logger.info('Image Number = {}, Image file = {}'.format(idx, img_path))
logger.info('Image Size = {} x {}'.format(img_width, img_height))
logger.info('Display Size = {} x {}'.format(random_ann.disp_width, random_ann.disp_height))
sc_factor_1 = (img_height * 1.) / random_ann.disp_height
sc_factor_2 = (img_width * 1.) / random_ann.disp_width
logger.info('Factor: {} {}'.format(sc_factor_1, sc_factor_2))
bbox = random_ann.bbox
bbox.x1 = bbox.x1 * sc_factor_1
bbox.x2 = bbox.x2 * sc_factor_1
bbox.y1 = bbox.y1 * sc_factor_1
bbox.y2 = bbox.y2 * sc_factor_1
if self._dbg:
self._data_fetched.append((img_path.name, bbox.x1, bbox.y1,
bbox.x2, bbox.y2))
# return image_io.image_to_tensor(image), bbox
return image, bbox
def make_training_examples(sample_gen):
"""
1. First decide the current search region, which is
kContextFactor(=2) * current bounding box.
2. Crop the valid search region and copy to the new padded image
3. Recenter the actual bounding box of the object to the new
padded image
4. Scale the bounding box for regression
"""
images = []
targets = []
bboxes = []
image, target, bbox_gt_scaled = sample_gen.make_true_sample()
images.append(image)
targets.append(target)
bboxes.append(bbox_gt_scaled)
# Generate more number of examples
images, targets, bbox_gt_scaled = sample_gen.make_training_samples(10, images, targets, bboxes)
if __name__ == "__main__":
imagnet_path = '/media/nthere/datasets/ISLVRC2014_Det/dummy/'
img_dir = Path(imagnet_path).joinpath('images')
ann_dir = Path(imagnet_path).joinpath('gt')
imagenetD = ImageNetDataset(str(img_dir), str(ann_dir),
isTrain=False, val_ratio=0, dbg=True)
for i, (image, bbox, image, bbox) in enumerate(imagenetD):
pass
sample_gen = sample_generator(5, 15, -0.4, 0.4, dbg=True, env=imagenetD._env)
for i, (image, bbox, image, bbox) in enumerate(imagenetD):
sample_gen.reset(bbox, bbox, image, image)
make_training_examples(sample_gen)