def transform_tr(self, sample):
composed_transforms = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.RandomScaleCrop(base_size=self.args.image_size, crop_size=self.args.crop_size),
tr.RandomGaussianBlur(),
tr.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
tr.ToTensor()])
return composed_transforms(sample)
def transform_tr(self, sample):
composed_transforms = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.RandomVerticalFlip(),
tr.RandomScaleCrop(base_size=self.args.base_size,
crop_size=self.args.crop_size,
fill=255),
tr.RandomRotate(),
tr.RandomGaussianBlur(),
tr.RandomNoise(),
# mean and std calculated by calculateMeanStd.py
tr.Normalize(mean=(0.656963, 0.621670, 0.550278),
std=(0.300198, 0.303201, 0.334976)),
tr.ToTensor()
])
return composed_transforms(sample)
def __init__(self,
args,
split,
inputRes,
augment=False,
transform=None,
target_transform=None):
self._year = args.year
self._phase = split
self.transform = transform
self.target_transform = target_transform
self.inputRes = inputRes
self.augment = augment
self.augment_transform = None
self._single_object = False
assert args.year == "2017" or args.year == "2016"
if augment:
self.augment_transform = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.ScaleNRotate(rots=(-args.rotation, args.rotation),
scales=(.75, 1.25))
])
self.imagefiles = []
self.maskfiles = []
self.flowfiles = []
self.edgefiles = []
self.hedfiles = []
if split == 'train':
self.load_davis(args)
self.load_youtubevos(args)
else:
self.load_davis(args)
def __init__(self, args, split, inputRes, augment=False,
transform=None, target_transform=None):
self._year = args.year
self._phase = split
self.transform = transform
self.target_transform = target_transform
self.inputRes = inputRes
self.augment = augment
self.augment_transform = None
self._single_object = False
assert args.year == "2017" or args.year == "2016"
if augment:
self.augment_transform = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.ScaleNRotate(rots=(-args.rotation, args.rotation),
scales=(.75, 1.25))])
self._db_sequences = db_read_sequences(args.year, self._phase)
# Check lmdb existance. If not proceed with standard dataloader.
lmdb_env_seq_dir = osp.join(cfg.PATH.DATA, 'lmdb_seq')
lmdb_env_annot_dir = osp.join(cfg.PATH.DATA, 'lmdb_annot')
if osp.isdir(lmdb_env_seq_dir) and osp.isdir(lmdb_env_annot_dir):
lmdb_env_seq = lmdb.open(lmdb_env_seq_dir)
lmdb_env_annot = lmdb.open(lmdb_env_annot_dir)
else:
lmdb_env_seq = None
lmdb_env_annot = None
print('LMDB not found. This could affect the data loading time. It is recommended to use LMDB.')
self.sequences = [Sequence(self._phase, s.name, lmdb_env=lmdb_env_seq) for s in self._db_sequences]
self._db_sequences = db_read_sequences(args.year, self._phase)
# Load annotations
self.annotations = [Annotation(self._phase,s.name, self._single_object, lmdb_env=lmdb_env_annot) for s in self._db_sequences]
self._db_sequences = db_read_sequences(args.year, self._phase)
# Load Videos
self.videos = []
for seq, s in zip(self.sequences, self._db_sequences):
if s['set'] == self._phase:
self.videos.append(s['name'])
self.imagefiles = []
self.maskfiles = []
self.flowfiles = []
self.edgefiles = []
for _video in self.videos:
imagefiles = sorted(glob.glob(os.path.join(cfg.PATH.SEQUENCES, _video, '*.jpg')))
maskfiles = sorted(glob.glob(os.path.join(cfg.PATH.ANNOTATIONS, _video, '*.png')))
flowfiles = sorted(glob.glob(os.path.join(cfg.PATH.FLOW, _video, '*.png')))
edgefiles = sorted(glob.glob(os.path.join(cfg.PATH.ANNOTATIONS_EDGE, _video, '*.png')))
self.imagefiles.extend(imagefiles[:-1])
self.maskfiles.extend(maskfiles[:-1])
self.flowfiles.extend(flowfiles)
self.edgefiles.extend(edgefiles[:-1])
print('images: ', len(self.imagefiles))
print('masks: ', len(self.maskfiles))
assert(len(self.imagefiles) == len(self.maskfiles) == len(self.flowfiles) == len(self.edgefiles))
joint_transform = joint_transforms.Compose([
joint_transforms.ImageResize(520),
joint_transforms.RandomCrop(473),
joint_transforms.RandomHorizontallyFlip(),
joint_transforms.RandomRotate(10)
])
img_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
target_transform = transforms.ToTensor()
davis_transforms = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.ScaleNRotate(rots=(-30, 30), scales=(.75, 1.25)),
tr.ToTensor()]
)
# train_set = ImageFolder(msra10k_path, joint_transform, img_transform, target_transform)
if args['train_loader'] == 'video_sequence':
train_set = VideoSequenceFolder(video_seq_path, video_seq_gt_path, imgs_file, joint_transform, img_transform, target_transform)
else:
# train_set = DAVIS2016(db_root_dir='/home/ty/data/davis', train=True, transform=None)
train_set = DAVIS_Single(db_image_dir='/home/ty/data/Pre-train', train=True, transform=None)
# train_set = VideoImageFolder(video_train_path, imgs_file, joint_transform, img_transform, target_transform)
train_loader = DataLoader(train_set, batch_size=args['train_batch_size'], num_workers=4, shuffle=True)
def __init__(self,
args,
transform=None,
target_transform=None,
augment=False,
split='train',
inputRes=None):
self._phase = split
self._single_object = args.single_object
self._length_clip = args.length_clip
self.transform = transform
self.target_transform = target_transform
self.split = split
self.inputRes = inputRes
self.max_seq_len = args.gt_maxseqlen
self.dataset = args.dataset
self.flip = augment
if augment:
self.augment_transform = transforms.Compose([
tr.RandomHorizontalFlip(),
tr.ScaleNRotate(rots=(-args.rotation, args.rotation),
scales=(.75, 1.25))
])
else:
self.augment_transform = None
if self._phase == phase.TRAIN.value:
self._db_sequences = db_read_sequences_train()
elif self._phase == phase.VAL.value:
self._db_sequences = db_read_sequences_val()
elif self._phase == phase.TRAINVAL.value:
self._db_sequences = db_read_sequences_trainval()
else: #self._phase == 'test':
self._db_sequences = db_read_sequences_test()
# Check lmdb existance. If not proceed with standard dataloader.
lmdb_env_seq_dir = osp.join(cfg.PATH.DATA, 'lmdb_seq')
lmdb_env_annot_dir = osp.join(cfg.PATH.DATA, 'lmdb_annot')
if osp.isdir(lmdb_env_seq_dir) and osp.isdir(lmdb_env_annot_dir):
lmdb_env_seq = lmdb.open(lmdb_env_seq_dir)
lmdb_env_annot = lmdb.open(lmdb_env_annot_dir)
else:
lmdb_env_seq = None
lmdb_env_annot = None
print(
'LMDB not found. This could affect the data loading time. It is recommended to use LMDB.'
)
# Load sequences
self.sequences = [
Sequence(self._phase, s, lmdb_env=lmdb_env_seq)
for s in self._db_sequences
]
# Load annotations
self.annotations = [
Annotation(self._phase,
s,
self._single_object,
lmdb_env=lmdb_env_annot) for s in self._db_sequences
]
# Load sequences
self.videos = []
for seq, s in zip(self.sequences, self._db_sequences):
self.videos.append(s)
self.imagefiles = []
self.maskfiles = []
for _video in self.videos:
imagefiles = sorted(
glob.glob(
os.path.join(cfg.PATH.SEQUENCES_TRAIN, _video, '*.jpg')))
maskfiles = sorted(
glob.glob(
os.path.join(cfg.PATH.ANNOTATIONS_TRAIN, _video, '*.png')))
self.imagefiles.extend(imagefiles)
self.maskfiles.extend(maskfiles)
print('images: ', len(self.imagefiles))
print('masks: ', len(self.maskfiles))