def task_datasets(task, vali_frac=0, colorspace='RGB'):
learn, test = next(task.xval_splits())
learn.tag = 'learn'
# Split everything in the learning set into training / validation
n_learn = len(learn)
n_vali = int(n_learn * vali_frac)
train = learn[n_vali:]
vali = learn[:n_vali]
vali.tag = 'vali'
train.tag = 'train'
if ub.argflag('--all'):
# HACK EVERYTHING TOGETHER
train = learn + test
from clab import inputs
vali = inputs.Inputs()
test = inputs.Inputs()
train_dataset = SSegInputsWrapper(train, task, colorspace=colorspace)
vali_dataset = SSegInputsWrapper(vali, task, colorspace=colorspace)
test_dataset = SSegInputsWrapper(test, task, colorspace=colorspace)
# train_dataset.augment = True
print('* len(train_dataset) = {}'.format(len(train_dataset)))
print('* len(vali_dataset) = {}'.format(len(vali_dataset)))
print('* len(test_dataset) = {}'.format(len(test_dataset)))
datasets = {
'train': train_dataset,
'vali': vali_dataset,
'test': test_dataset,
}
return datasets
def load_task_dataset(taskname,
vali_frac=0,
colorspace='RGB',
combine=None,
boundary=True,
arch=None,
halfcombo=None):
task = get_task(taskname, boundary=boundary, arch=arch)
learn, test = next(task.xval_splits())
learn.tag = 'learn'
# Split everything in the learning set into training / validation
n_learn = len(learn)
n_vali = int(n_learn * vali_frac)
train = learn[n_vali:]
vali = learn[:n_vali]
vali.tag = 'vali'
train.tag = 'train'
if combine is None:
combine = ub.argflag('--combine')
if halfcombo is None:
halfcombo = ub.argflag('--halfcombo')
if halfcombo:
# decrease testing for training
n = len(test) // 2
new_test = test[n:]
train = learn + test[:n]
test = new_test
train.tag = 'train_h'
test.tag = 'test_h'
if combine:
# HACK EVERYTHING TOGETHER
train = learn + test
from clab import inputs
vali = inputs.Inputs()
test = inputs.Inputs()
train_dataset = SSegInputsWrapper(train, task, colorspace=colorspace)
vali_dataset = SSegInputsWrapper(vali, task, colorspace=colorspace)
test_dataset = SSegInputsWrapper(test, task, colorspace=colorspace)
# train_dataset.augment = True
print('* len(train_dataset) = {}'.format(len(train_dataset)))
print('* len(vali_dataset) = {}'.format(len(vali_dataset)))
print('* len(test_dataset) = {}'.format(len(test_dataset)))
datasets = {
'train': train_dataset,
'vali': vali_dataset,
'test': test_dataset,
}
return datasets
def _mode_new_input(prep, mode, input, clear=False, mult=1):
out_dpaths = prep._mode_paths(mode, input, clear=clear)
new_input = inputs.Inputs()
new_input.tag = mode
if 'im' in out_dpaths:
new_input.imdir = out_dpaths['im']
if 'gt' in out_dpaths:
new_input.gtdir = out_dpaths['gt']
if 'aux' in out_dpaths:
new_input.auxdir = out_dpaths['aux']
if not clear:
try:
new_input.prepare_image_paths()
except AssertionError: # hack
return prep._mode_new_input(mode, input, clear=True)
if len(new_input.paths) > 0:
n_loaded = min(map(len, new_input.paths.values()))
min_n_expected = len(input) * mult
print(' * n_loaded = {!r}'.format(n_loaded))
print(' * min_n_expected = {!r}'.format(min_n_expected))
# Short curcuit augmentation if we found stuff
if n_loaded >= min_n_expected:
print('short circuit {}'.format(mode))
return new_input, True
if 'im' in out_dpaths:
new_input.im_paths = []
if 'gt' in out_dpaths:
new_input.gt_paths = []
if 'aux' in out_dpaths:
new_input.aux_paths = {k: [] for k in input.aux_paths.keys()}
return new_input, False