def get_dataset(dataset_root, batch_size, net_size):
subset = [('2007', 'test')]
datatransform = transform.YOLO3DefaultValTransform(mean=(0, 0, 0),
std=(1, 1, 1))
valset = VOCdataset(dataset_root,
datatransform,
subset,
batch_size,
net_size,
shuffle=False)
valset_iter = tf.data.Dataset.from_generator(
valset, ((tf.float32, tf.string, tf.string, tf.float32, tf.float32,
tf.float32, tf.float32, tf.float32)))
valset_iter = valset_iter.batch(1).prefetch(tf.data.experimental.AUTOTUNE)
subset = [('2007', 'trainval'), ('2012', 'trainval')]
datatransform = transform.YOLO3DefaultTrainTransform(mean=(0, 0, 0),
std=(1, 1, 1))
trainset = VOCdataset(dataset_root,
datatransform,
subset,
batch_size,
net_size,
shuffle=True)
trainset_iter = tf.data.Dataset.from_generator(
trainset, ((tf.float32, tf.string, tf.string, tf.float32, tf.float32,
tf.float32, tf.float32, tf.float32)))
# be careful to drop the last smaller batch if using tf.function
trainset_iter = trainset_iter.batch(1, drop_remainder=True).prefetch(
tf.data.experimental.AUTOTUNE)
return trainset_iter, valset_iter
def get_dataset(dataset_root, batch_size, net_size):
subset = [('2007', 'test')]
datatransform = transform.YOLO3DefaultValTransform(mean=(0, 0, 0),
std=(1, 1, 1))
valset = VOCdataset(dataset_root,
datatransform,
subset,
batch_size,
net_size,
istrain=False)
valset = DataLoader(dataset=valset,
batch_size=1,
shuffle=False,
num_workers=1,
pin_memory=True)
subset = [('2007', 'trainval'), ('2012', 'trainval')]
datatransform = transform.YOLO3DefaultTrainTransform(mean=(0, 0, 0),
std=(1, 1, 1))
trainset = VOCdataset(dataset_root,
datatransform,
subset,
batch_size,
net_size,
istrain=True)
trainset = DataLoader(dataset=trainset,
batch_size=1,
shuffle=True,
num_workers=1,
pin_memory=True)
return trainset, valset
def get_dataset(config):
config['subset'] = 'val'
datatransform = transform.YOLO3DefaultValTransform(height=416,
width=416,
mean=(0, 0, 0),
std=(1, 1, 1))
valset = PascalDataset(config, datatransform)
generator = DataGenerator(valset)
valset = tf.data.Dataset.from_generator(
generator, ((tf.float32, tf.string, tf.float32, tf.float32, tf.float32,
tf.float32, tf.float32)))
valset = valset.batch(config['batch_size']).prefetch(
tf.data.experimental.AUTOTUNE)
config["subset"] = 'train'
datatransform = transform.YOLO3DefaultTrainTransform(height=416,
width=416,
mean=(0, 0, 0),
std=(1, 1, 1))
trainset = PascalDataset(config, datatransform)
generator = DataGenerator(trainset, shuffle=True)
trainset = tf.data.Dataset.from_generator(
generator, ((tf.float32, tf.string, tf.float32, tf.float32, tf.float32,
tf.float32, tf.float32)))
#be careful to drop the last smaller batch if using tf.function
trainset = trainset.batch(config['batch_size'],
drop_remainder=True).prefetch(
tf.data.experimental.AUTOTUNE)
return trainset, valset
def get_dataset(dataset_root, batch_size, net_size): datatransform = transform.YOLO3DefaultValTransform(mean=(0, 0, 0), std=(1, 1, 1)) valset = COCOdataset(dataset_root, datatransform, subset='val', batchsize=batch_size, netsize=net_size,istrain=False) valset = torch.utils.data.DataLoader(dataset=valset,batch_size=1,shuffle=False,num_workers=4,pin_memory=True) trainset = COCOdataset(dataset_root, datatransform, subset='train', batchsize=batch_size, netsize=net_size,istrain=True) trainset = torch.utils.data.DataLoader(dataset=trainset,batch_size=1,shuffle=True,num_workers=4,pin_memory=True) return trainset, valset
def get_dataset(dataset_root, batch_size, trainsizes, testsize, numworkers=4, debug=False, gt_pergrid=3):
subset = [('2007', 'trainval'), ('2012', 'trainval')]
datatransform = transform.YOLO3DefaultTrainTransform(mean=(0, 0, 0), std=(1, 1, 1))
trainset = VOCdataset(dataset_root, datatransform, subset, batch_size, trainsizes=trainsizes, testsize=testsize,
istrain=True, gt_pergrid=gt_pergrid)
trainset = DataLoader(dataset=trainset, batch_size=1, shuffle=False, num_workers=numworkers, pin_memory=True)
subset = [('2007', 'test')]
datatransform = transform.YOLO3DefaultValTransform(mean=(0, 0, 0), std=(1, 1, 1))
valset = VOCdataset(dataset_root, datatransform, subset, batch_size, trainsizes=trainsizes, testsize=testsize,
istrain=False, gt_pergrid=gt_pergrid)
valset = DataLoader(dataset=valset, batch_size=1, shuffle=False, num_workers=numworkers, pin_memory=True)
return trainset, valset
def get_dataset(dataset_root,batch_size,net_size):
datatransform = transform.YOLO3DefaultValTransform(mean=(0, 0, 0), std=(1, 1, 1))
valset = COCOdataset(dataset_root,datatransform,subset='val',shuffle=False,batchsize=batch_size,netsize=net_size)
valset = tf.data.Dataset.from_generator(valset,
((tf.float32, tf.string,tf.string, tf.float32, tf.float32, tf.float32, tf.float32,
tf.float32)))
valset = valset.batch(1).prefetch(tf.data.experimental.AUTOTUNE)
datatransform = transform.YOLO3DefaultTrainTransform(mean=(0, 0, 0), std=(1, 1, 1))
trainset = COCOdataset(dataset_root,datatransform,subset='train',shuffle=True,batchsize=batch_size,netsize=net_size)
trainset = tf.data.Dataset.from_generator(trainset,
((tf.float32, tf.string,tf.string, tf.float32, tf.float32, tf.float32, tf.float32,
tf.float32)))
#be careful to drop the last smaller batch if using tf.function
trainset = trainset.batch(1,drop_remainder=True).prefetch(tf.data.experimental.AUTOTUNE)
return trainset, valset
def get_dataset(dataset_root,
batch_size,
trainsizes,
testsize,
numworkers=4,
debug=False,
gt_pergrid=3):
datatransform = transform.YOLO3DefaultValTransform(mean=(0, 0, 0),
std=(1, 1, 1))
valset = COCOdataset(dataset_root,
datatransform,
subset='val',
batchsize=batch_size,
trainsizes=trainsizes,
testsize=testsize,
istrain=False,
gt_pergrid=gt_pergrid)
valset = torch.utils.data.DataLoader(dataset=valset,
batch_size=1,
shuffle=False,
num_workers=numworkers,
pin_memory=True)
if debug:
return valset, valset
trainset = COCOdataset(dataset_root,
datatransform,
subset='train',
batchsize=batch_size,
trainsizes=trainsizes,
testsize=testsize,
istrain=True,
gt_pergrid=gt_pergrid)
trainset = torch.utils.data.DataLoader(dataset=trainset,
batch_size=1,
shuffle=True,
num_workers=numworkers,
pin_memory=True)
return trainset, valset