def __init__(self, img_size, range01=False, rgb_order=False, dummy=False):
images_dir = settings.get_data_dir('office_webcam')
super(OfficeWebcamDataset, self).__init__(img_size,
range01,
rgb_order,
images_dir,
dummy=dummy)
def __init__(self, img_size, range01=False, rgb_order=False, dummy=False):
images_dir = settings.get_data_dir('amazon_dslr')
super(OfficeAmazon2DslrDataset, self).__init__(img_size,
range01,
rgb_order,
images_dir,
dummy=dummy)
def __init__(self,
img_size,
range01=False,
rgb_order=False,
dummy=False,
download=True):
self.file_id = '0B4IapRTv9pJ1WGZVd1VDMmhwdlE'
self.filename = "domain_adaptation_images.tar.gz"
# download dataset.
if download:
download_file_from_google_drive(
self.file_id, os.path.join('./dataset/', self.filename))
if not _check_exists(os.path.join('./dataset/', self.filename)):
raise RuntimeError("Dataset not found." +
" You can use download=True to download it")
load_samples(os.path.join('./dataset/', self.filename))
images_dir = settings.get_data_dir('office_amazon')
super(OfficeAmazonDataset, self).__init__(img_size,
range01,
rgb_order,
images_dir,
dummy=dummy)
def __init__(self, img_size, range01=False, rgb_order=False, dummy=False):
test_dir = settings.get_data_dir('visda17_clf_test')
file_list_path = os.path.join(test_dir, 'image_list.txt')
super(TestDataset, self).__init__(img_size,
range01,
rgb_order,
file_list_path,
test_dir,
has_ground_truth=False,
dummy=dummy)
def __init__(self, img_size, range01=False, rgb_order=False, dummy=False):
val_dir = settings.get_data_dir('visda17_clf_validation')
file_list_path = os.path.join(val_dir, 'image_list.txt')
super(ValidationDataset, self).__init__(img_size,
range01,
rgb_order,
file_list_path,
val_dir,
has_ground_truth=True,
dummy=dummy)
def __init__(self, img_size, range01=False, rgb_order=False, dummy=False):
train_dir = settings.get_data_dir('visda17_clf_train')
file_list_path = os.path.join(train_dir, 'image_list.txt')
super(TrainDataset, self).__init__(img_size,
range01,
rgb_order,
file_list_path,
train_dir,
has_ground_truth=True,
dummy=dummy)
self.object_ids = []
self.cam_yaw = []
self.light_yaw = []
self.cam_pitch = []
self.obj_id_to_idx = {}
self.cam_yaw_to_idx = {}
self.light_yaw_to_idx = {}
self.cam_pitch_to_idx = {}
for sample_idx, name in enumerate(self.names):
fn, _ = os.path.splitext(name)
object_id, _, tail = fn.partition('__')
c_yaw, l_yaw, c_pitch = tail.split('_')
c_yaw = float(c_yaw)
l_yaw = float(l_yaw)
c_pitch = float(c_pitch)
obj_id_idx = self.obj_id_to_idx.setdefault(object_id,
len(self.obj_id_to_idx))
c_yaw_idx = self.cam_yaw_to_idx.setdefault(
c_yaw, len(self.cam_yaw_to_idx))
l_yaw_idx = self.light_yaw_to_idx.setdefault(
l_yaw, len(self.light_yaw_to_idx))
c_pitch_idx = self.cam_pitch_to_idx.setdefault(
c_pitch, len(self.cam_pitch_to_idx))
self.object_ids.append(obj_id_idx)
self.cam_yaw.append(c_yaw_idx)
self.light_yaw.append(l_yaw_idx)
self.cam_pitch.append(c_pitch_idx)
self.object_ids = np.array(self.object_ids, dtype=np.int32)
self.cam_yaw = np.array(self.cam_yaw, dtype=np.int32)
self.light_yaw = np.array(self.light_yaw, dtype=np.int32)
self.cam_pitch = np.array(self.cam_pitch, dtype=np.int32)
sample_ndxs = np.arange(len(self.object_ids))
self.samples_by_obj_id = [
sample_ndxs[self.object_ids == i]
for i in range(len(self.obj_id_to_idx))
]
self.samples_by_cam_yaw = [
sample_ndxs[self.cam_yaw == i]
for i in range(len(self.cam_yaw_to_idx))
]
self.samples_by_light_yaw = [
sample_ndxs[self.light_yaw == i]
for i in range(len(self.light_yaw_to_idx))
]
self.samples_by_cam_pitch = [
sample_ndxs[self.cam_pitch == i]
for i in range(len(self.cam_pitch_to_idx))
]
self.obj_X = self.ObjectImageAccessor(self)