def __init__(self,
video_path,
patch_size,
window_len,
rotate=10,
scale=1.2,
full_size=640,
is_train=True):
super(VidListv2, self).__init__()
self.data_dir = video_path
self.window_len = window_len
normalize = transforms.Normalize(mean=(128, 128, 128),
std=(128, 128, 128))
self.transforms1 = transforms.Compose([
transforms.RandomRotate(rotate),
transforms.ResizeandPad(full_size),
transforms.RandomCrop(patch_size),
transforms.ToTensor(), normalize
])
self.transforms2 = transforms.Compose([
transforms.ResizeandPad(full_size),
transforms.ToTensor(), normalize
])
self.is_train = is_train
self.list = list_sequences(video_path, set_ids=list(
range(12))) # training sets: 0~11
def __init__(self,
video_path,
list_path,
patch_size,
window_len,
rotate=10,
scale=1.2,
full_size=640,
is_train=True):
super(VidListv2, self).__init__()
self.data_dir = video_path
self.list_path = list_path
self.window_len = window_len
normalize = transforms.Normalize(mean=(128, 128, 128),
std=(128, 128, 128))
self.transforms1 = transforms.Compose([
transforms.RandomRotate(rotate),
# transforms.RandomScale(scale),
transforms.ResizeandPad(full_size),
transforms.RandomCrop(patch_size),
transforms.ToTensor(),
normalize
])
self.transforms2 = transforms.Compose([
transforms.ResizeandPad(full_size),
transforms.ToTensor(), normalize
])
self.is_train = is_train
self.read_list()
def __init__(self,
video_path,
list_path,
patch_size,
rotate=10,
scale=1.2,
is_train=True,
moreaug=True):
super(VidListv1, self).__init__()
self.data_dir = video_path
self.list_path = list_path
normalize = transforms.Normalize(mean=(128, 128, 128),
std=(128, 128, 128))
t = []
if rotate > 0:
t.append(transforms.RandomRotate(rotate))
if scale > 0:
t.append(transforms.RandomScale(scale))
t.extend([
transforms.RandomCrop(patch_size, seperate=moreaug),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(), normalize
])
self.transforms = transforms.Compose(t)
self.is_train = is_train
self.read_list()
def __init__(self,
video_path,
list_path,
patch_size,
rotate=10,
scale=1.2,
is_train=True,
moreaug=True):
super(VidListv1, self).__init__()
csv_path = "/raid/codes/CorrFlows/functional/feeder/dataset/oxuva.csv"
filenames = open(csv_path).readlines()
frame_all = [filename.split(',')[0].strip() for filename in filenames]
nframes = [
int(filename.split(',')[1].strip()) for filename in filenames
]
self.data_dir = video_path
self.list = frame_all
self.nframes = nframes
normalize = transforms.Normalize(mean=(128, 128, 128),
std=(128, 128, 128))
t = []
if rotate > 0:
t.append(transforms.RandomRotate(rotate))
if scale > 0:
t.append(transforms.RandomScale(scale))
t.extend([
transforms.RandomCrop(patch_size, seperate=moreaug),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(), normalize
])
self.transforms = transforms.Compose(t)
self.is_train = is_train
def __init__(self,
video_path,
list_path,
patch_size,
window_len,
rotate=10,
scale=1.2,
full_size=640,
is_train=True):
super(VidListv2, self).__init__()
csv_path = "/raid/codes/UVC/libs/GOT-new.csv"
filenames = open(csv_path).readlines()
frame_all = [filename.split(',')[0].strip() for filename in filenames]
nframes = [
int(filename.split(',')[1].strip()) for filename in filenames
]
self.data_dir = video_path
self.list = frame_all
self.window_len = window_len
self.nframes = nframes
normalize = transforms.Normalize(mean=(128, 128, 128),
std=(128, 128, 128))
self.transforms1 = transforms.Compose([
transforms.RandomRotate(rotate),
# transforms.RandomScale(scale),
transforms.ResizeandPad_1(full_size),
transforms.RandomCrop(patch_size),
transforms.ToTensor(),
normalize
])
self.transforms2 = transforms.Compose([
transforms.ResizeandPad_1(full_size),
transforms.ToTensor(), normalize
])
self.is_train = is_train
self.video_list = []
for filename in filenames:
record = VideoRecord()
record.path = os.path.join(video_path,
filename.split(',')[0].strip())
record.num_frames = int(filename.split(',')[1].strip(
)) #len(glob.glob(os.path.join(video_path, '*.jpg')))
record.label = filename.split(',')[0].strip()
self.video_list.append(record)
def __init__(self, video_path, list_path, patch_size, window_len, rotate = 10, scale = 1.2, full_size = 640, is_train=True):
super(VidListv2, self).__init__()
csv_path = "/raid/codes/CorrFlows/GOT_10k.csv"
filenames = open(csv_path).readlines()
frame_all = [filename.split(',')[0].strip() for filename in filenames]
nframes = [int(filename.split(',')[1].strip()) for filename in filenames]
self.data_dir = video_path
self.list = frame_all
self.window_len = window_len
self.nframes = nframes
normalize = transforms.Normalize(mean = (128, 128, 128), std = (128, 128, 128))
self.transforms1 = transforms.Compose([
transforms.RandomRotate(rotate),
# transforms.RandomScale(scale),
transforms.ResizeandPad(full_size),
transforms.RandomCrop(patch_size),
transforms.ToTensor(),
normalize])
self.transforms2 = transforms.Compose([
transforms.ResizeandPad(full_size),
transforms.ToTensor(),
normalize])
self.is_train = is_train
def __init__(self,
video_path,
list_path,
patch_size,
rotate=10,
scale=1.2,
is_train=True,
moreaug=True):
super(VidListv1, self).__init__()
csv_path = "/raid/codes/UVC/libs/GOT-new.csv"
filenames = open(csv_path).readlines()
frame_all = [filename.split(',')[0].strip() for filename in filenames]
nframes = [
int(filename.split(',')[1].strip()) for filename in filenames
]
self.data_dir = video_path
self.list = frame_all
self.nframes = nframes
normalize = transforms.Normalize(mean=(128, 128, 128),
std=(128, 128, 128))
t = []
if rotate > 0:
t.append(transforms.RandomRotate(rotate))
if scale > 0:
t.append(transforms.RandomScale(scale))
t.extend([
transforms.RandomCrop(patch_size, seperate=moreaug),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(), normalize
])
self.transforms = transforms.Compose(t)
self.is_train = is_train
self.video_list = []
for filename in filenames:
record = VideoRecord()
record.path = os.path.join(video_path,
filename.split(',')[0].strip())
record.num_frames = int(filename.split(',')[1].strip(
)) # len(glob.glob(os.path.join(video_path, '*.jpg')))
record.label = filename.split(',')[0].strip()
self.video_list.append(record)
root = path.partition("Kinetices/")[0]
if not exists(path):
raise Exception(
"{} does not exist in kinet_dataset.py.".format(path))
self.list = [
line.replace("/Data/", root).strip() for line in open(path, 'r')
]
if __name__ == '__main__':
normalize = transforms.Normalize(mean=(128, 128, 128), std=(128, 128, 128))
t = []
t.extend([
transforms.RandomCrop(256),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(), normalize
])
dataset_train = VidList('/home/xtli/DATA/compress/train_256/',
'/home/xtli/DATA/compress/train.txt',
transforms.Compose(t),
window_len=2)
train_loader = torch.utils.data.DataLoader(dataset_train,
batch_size=16,
shuffle=True,
num_workers=8,
drop_last=True)
start_time = time.time()
for i, (frames) in enumerate(train_loader):
print(i)
def create_transforms(crop_size):
normalize = transforms.Normalize(mean=(128, 128, 128), std=(128, 128, 128))
t = []
t.extend([transforms.ToTensor(), normalize])
return transforms.Compose(t)
def read_list(self):
path = join(self.list_path)
root = path.partition("Kinetices/")[0]
if not exists(path):
raise Exception("{} does not exist in kinet_dataset.py.".format(path))
self.list = [line.replace("/Data/", root).strip() for line in open(path, 'r')]
if __name__ == '__main__':
normalize = transforms.Normalize(mean = (128, 128, 128),
std = (128, 128, 128))
t = []
t.extend([transforms.RandomCrop(256),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize])
dataset_train = VidList('/home/xtli/DATA/compress/train_256/',
'/home/xtli/DATA/compress/train.txt',
transforms.Compose(t), window_len=2)
train_loader = torch.utils.data.DataLoader(dataset_train,
batch_size = 16,
shuffle = True,
num_workers=8,
drop_last=True)
start_time = time.time()
for i, (frames) in enumerate(train_loader):
print(i)
if(i >= 1000):