def transform_val(self, sample):
composed_transforms = transforms.Compose([
tr.FixScaleCrop(crop_size=self.args.crop_size),
tr.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
tr.ToTensor()])
return composed_transforms(sample)
def transform_val(self, sample):
composed_transforms = transforms.Compose([
# tr.Resize(size=(64, 64)),
tr.PaddingSurround(),
tr.UpperThreshold(0),
tr.Normalize(),
tr.ToTensor()
])
return composed_transforms(sample)
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_val(self, sample):
composed_transforms = transforms.Compose([
tr.FixScaleCrop(crop_size=self.args.crop_size),
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, data_root, num_sample=3, Training=False):
self.Training = Training
self.augment_transform = None
self._single_object = False
self.num_sample = num_sample
if Training:
self.data_dir = os.path.join(data_root, 'Train')
self.augment_transform = transforms.Compose([
tr.Resize(scales=(512, 512)),
tr.Crop(scale=0.5),
tr.ToTensor()
])
else:
self.data_dir = os.path.join(data_root, 'Test')
self.augment_transform = transforms.Compose(
[tr.Resize(scales=(512, 512)),
tr.ToTensor()])
# Load Videos
self.videos = []
for seq in sorted(os.listdir(self.data_dir)):
self.videos.append(seq)
if Training:
random.shuffle(self.videos)
self.videoindex = {}
self.imagefiles = []
self.videofiles = []
offset = 0
for _video in self.videos:
imagefiles = sorted(
glob.glob(os.path.join(self.data_dir, _video, '*.jpg')))
self.imagefiles.extend(imagefiles)
self.videofiles.extend([_video] * len(imagefiles))
self.videoindex[_video] = [offset, offset + len(imagefiles)]
offset += len(imagefiles)
print('total video: ', len(self.videos))
print('total image: ', len(self.imagefiles))
def transform_val(self, sample):
"""
You can change transforms with <dataloader.custom_transforms>.
"""
composed_transforms = transforms.Compose([
tr.Resize(size=(64, 64)),
tr.Normalize(mean=(0.30273438, 0.30273438, 0.30273438),
std=(0.44050565, 0.44050565, 0.44050565)),
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)
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)
criterion = nn.BCEWithLogitsLoss().cuda()
}, {
'params':
[param for name, param in net.named_parameters() if name[-4:] != 'bias'],
'lr':
lr,
'weight_decay':
wd
}],
momentum=0.9)
# Preparation of the data loaders
# Define augmentation transformations as a composition
composed_transforms = transforms.Compose([
tr.RandomHorizontalFlip(),
# tr.ScaleNRotate(rots=(-30, 30), scales=(.75, 1.25)),
tr.ToTensor()
])
# Training dataset and its iterator
db_train = db.DAVIS2016(train=True,
db_root_dir=db_root_dir,
transform=composed_transforms,
seq_name=seq_name)
trainloader = DataLoader(db_train,
batch_size=p['trainBatch'],
shuffle=True,
num_workers=1)
# Testing dataset and its iterator
db_test = db.DAVIS2016(train=False,
db_root_dir=db_root_dir,
transform=tr.ToTensor(),
