def __init__(self, image_set, data_path, mc):
imdb.__init__(self, 'kitti_' + image_set, mc)
self._image_set = image_set
self._data_root_path = data_path
self._image_path = os.path.join(self._data_root_path, 'training',
'image_2')
self._label_path = os.path.join(self._data_root_path, 'training',
'label_2')
self._classes = self.mc.CLASS_NAMES
self._class_to_idx = dict(zip(self.classes, range(self.num_classes)))
# a list of string indices of images in the directory
self._image_idx = self._load_image_set_idx()
# a dict of image_idx -> [[cx, cy, w, h, cls_idx]]. x,y,w,h are not divided by
# the image width and height
self._rois = self._load_kitti_annotation()
## batch reader ##
self._perm_idx = None
self._cur_idx = 0
# TODO(bichen): add a random seed as parameter
self._shuffle_image_idx()
self._eval_tool = './src/dataset/kitti-eval/cpp/evaluate_object'
def __init__(self, image_set, data_path, mc):
imdb.__init__(self, 'vid_' + image_set, mc)
self._image_set = image_set
self._data_root_path = data_path
self._data_path = os.path.join(data_path, 'Data/VID', image_set)
self._idx_path = os.path.join(data_path, 'ImageSets/VID')
self._anno_path = os.path.join(data_path, 'Annotations/VID', image_set)
self._classes = self.mc.CLASS_NAMES
self._raw_cnames = ('n02691156', 'n02419796', 'n02131653', 'n02834778',
'n01503061', 'n02924116', 'n02958343', 'n02402425',
'n02084071', 'n02121808', 'n02503517', 'n02118333',
'n02510455', 'n02342885', 'n02374451', 'n02129165',
'n01674464', 'n02484322', 'n03790512', 'n02324045',
'n02509815', 'n02411705', 'n01726692', 'n02355227',
'n02129604', 'n04468005', 'n01662784', 'n04530566',
'n02062744', 'n02391049')
self._class_to_idx = dict(zip(self.classes, xrange(self.num_classes)))
self._raw_cname_to_idx = dict(
zip(self._raw_cnames, xrange(len(self._raw_cnames))))
# a list of string indices of images in the directory
self._image_idx = self._load_image_set_idx()
# a dict of image_idx -> [[cx, cy, w, h, cls_idx]]. x,y,w,h are not divided by
# the image width and height
self._rois = self._load_vid_annotation()
## batch reader ##
self._perm_idx = None
self._cur_idx = 0
# TODO(bichen): add a random seed as parameter
self._shuffle_image_idx()
def __init__(self, image_set, data_argument=True):
'''Initialize the class.
Args:
image_set: mark the property of data, value is 'train' or 'test'.
'''
imdb.__init__(self, 'kitti_' + image_set)
self._image_set = image_set
self._data_path = os.path.join(cfg.DATA_DIR, image_set)
self._classes = ['Car', 'Pedestrian', 'Cyclist']
self._dont_care = [
'DontCare', 'Misc', 'Person_sitting', 'Truck', 'Tram'
]
self._class_to_ind = dict(
zip(self._classes, xrange(len(self._classes))))
self._image_names, self._boxes, self._obj_num = self.read_annotations()
self._shuffle_inds()
# data_argument: if True, then apply data argument
# screen: if True, means the data has been preprocessed, then
# the bounding boxes info are [cls_index, xc, yc, w, h], and have been
# normalized
# image_ext: 'jpg' or 'png'
self.config = {
'data_argument': data_argument,
'screen': True,
'image_ext': '.png'
}
def __init__(self, image_set, year=2007, data_argument=True):
'''Initialize the class
By default: Using 2007 dataset
Args:
image_set: mark the property of data, value is 'train' or 'test'
'''
imdb.__init__(self, 'voc_' + year + '_' + image_set)
self._image_set = image_set
self._classes = ('aeroplane', 'bicycle', 'bird', 'boat', 'bottle',
'bus', 'car', 'cat', 'chair', 'cow', 'diningtable',
'dog', 'horse', 'motorbike', 'person', 'pottdplant',
'sheep', 'sofa', 'train', 'tvmonitor')
self._class_to_ind = dict(
zip(self._classes, xrange(len(self._classes))))
self._data_path = os.path.join(cfg.DATA_DIR, 'VOCdevkit' + year,
'VOC' + year)
# image name (without ext)
self._image_index = self._load_image_set_index()
self._image_names, self._boxes, self._obj_num = self.read_annotations()
self._shuffle_inds()
# data_argument: if True, then implement data argument
# image_ext: 'jpg' or 'png'
self.config = {'data_argument': data_argument, 'image_ext': '.jpg'}
def __init__(self, image_set, data_path, mc):
imdb.__init__(self, 'ILSVRC2013_' + image_set, mc)
self._image_set = image_set
self._data_path = data_path
self._classes = self.mc.CLASS_NAMES
#self._class_to_idx = dict(zip(self.classes, range(self.num_classes)))
# a list of string indices of images in the directory
self._image_idx = self._load_image_set_idx()
# a list of class labels
self.labels = self._load_imagenet_labels()
## batch reader ##
self._perm_idx = None
self._cur_idx = 0
self._shuffle_image_idx()
def __init__(self,
image_set='train',
data_path='/home/bryant/MATLAB-tools/caltech/',
mc=None):
imdb.__init__(self, 'caltech_' + image_set, mc)
self._image_set = image_set
self._data_root_path = data_path
self._image_path = os.path.join(self._data_root_path, 'raw_image/')
self._label_path = os.path.join(self._data_root_path,
'raw_annotations/')
if image_set == 'test':
self._image_set_num = [
'set06/', 'set07/', 'set08/', 'set09/', 'set10/'
]
self._image_set_stride = 30 # sample 1 frame every 4 frames from _image_set
elif image_set == 'valdation':
self._image_set_num = ['set05/']
self._image_set_stride = 30 # sample 1 frame every 4 frames from _image_set
elif image_set == 'train':
self._image_set_num = [
'set00/', 'set01/', 'set02/', 'set03/', 'set04/', 'set05/'
]
self._image_set_stride = 3 # sample 1 frame every 4 frames from _image_set
self._classes = self.mc.CLASS_NAMES
self._class_to_idx = dict(zip(self.classes, xrange(self.num_classes)))
# a list of string indices of images in the directory
self._image_idx = self._load_image_set_idx()
# a dict of image_idx -> [[cx, cy, w, h, cls_idx]]. x,y,w,h are not divided by
# the image width and height
if image_set == 'train':
self._rois = self.gt_roidb()
if mc.DATA_AUGMENTATION:
self._append_flip_rois()
## batch reader ##
self._perm_idx = None
self._cur_idx = 0
# TODO(bichen): add a random seed as parameter
if image_set == 'train':
self._shuffle_image_idx()
self._eval_tool = None # './src/dataset/kitti-eval/cpp/evaluate_object'
def __init__(self, image_set, year, data_path, mc):
imdb.__init__(self, 'voc_'+year+'_'+image_set, mc)
self._year = year
self._image_set = image_set
self._data_root_path = data_path
self._data_path = os.path.join(self._data_root_path, 'VOC' + self._year)
self._classes = self.mc.CLASS_NAMES
self._class_to_idx = dict(zip(self.classes, range(self.num_classes)))
# a list of string indices of images in the directory
self._image_idx = self._load_image_set_idx()
# a dict of image_idx -> [[cx, cy, w, h, cls_idx]]. x,y,w,h are not divided by
# the image width and height
self._rois = self._load_pascal_annotation()
## batch reader ##
self._perm_idx = None
self._cur_idx = 0
# TODO(bichen): add a random seed as parameter
self._shuffle_image_idx()
def __init__(self, name, mc):
imdb.__init__(self, name, mc)
self.left_margin = 0
self.right_margin = 0
self.top_margin = 0
self.bottom_margin = 0
def __init__(self, image_set, data_path, mc):
imdb.__init__(self, 'bosch' + image_set, mc)
self._image_set = image_set # train, test or trainval
self._data_root_path = data_path
#self._image_path = os.path.join(self._data_root_path, 'training', 'image_2')
#self._label_path = os.path.join(self._data_root_path, 'training', 'label_2')
self._classes = self.mc.CLASS_NAMES
self._class_to_idx = dict(zip(self.classes, range(self.num_classes)))
self._input_yaml = image_set + ".yaml"
self._bosch_data = yaml.load(
open(os.path.join(data_path, self._input_yaml), 'rb').read())
#image_dict= bosch_data[0] # bosch_data content image dictionary 'boxes' and 'path'
# a list of string indices of images in the directory
self._image_idx = []
self._image_path = []
# a dict of image_idx -> [[cx, cy, w, h, cls_idx]]. x,y,w,h are not divided by
# the image width and height
# '006303': [[75.735, 243.735, 151.47, 127.01000000000002, 0], [111.485, 238.64499999999998, 222.97, 114.95000000000002, 0], [237.58, 216.33999999999997, 116.94, 64.84, 0], [303.685, 210.755, 151.11, 58.44999999999999, 0], [434.94, 197.68, 52.72000000000003, 37.78, 0], ]
self._rois = {}
idx_counter = 0
for imgdict in self._bosch_data:
image_path = os.path.join(data_path, imgdict['path'])
roi = {}
bboxes = []
for box in imgdict['boxes']:
w = box["x_max"] - box["x_min"]
h = box["y_max"] - box["y_min"]
if w < 0:
print("Warning negative value of width %f, %s" %
(w, imgdict['path']))
w = abs(w)
if h < 0:
print("Warning negative value of height %f, %s" %
(h, imgdict['path']))
h = abs(h)
cx = box["x_min"] + w / 2
cy = box["y_min"] + h / 2
class_idx = 3
if "Red" in box['label']:
class_idx = 0
elif 'Yellow' in box['label']:
class_idx = 1
elif 'Green' in box['label']:
class_idx = 2
bboxes.append([cx, cy, w, h, class_idx])
self._image_idx.append(idx_counter)
self._rois[idx_counter] = bboxes
self._image_path.append(image_path)
idx_counter = idx_counter + 1
## batch reader ##
self._perm_idx = None
self._cur_idx = 0
self._shuffle_image_idx()
