def data_prep_function_train(x,
p_transform=p_transform,
p_augmentation=p_augmentation,
**kwargs):
x = x.convert('RGB')
x = np.array(x)
x = imresize(x, (128, 128, 3), interp='lanczos')
x = np.swapaxes(x, 0, 2)
x = x / 255.
x -= mean
x /= std
x = x.astype(np.float32)
pert_aug = dict((k, p_augmentation[k])
for k in ('zoom_range', 'rotation_range', 'shear_range',
'translation_range', 'do_flip', 'allow_stretch')
if k in p_augmentation)
x = data_transforms.perturb(x,
pert_aug,
p_transform['patch_size'],
rng,
n_channels=p_transform["channels"])
losless_aug = dict((k, p_augmentation[k]) for k in ('rot90_values', 'flip')
if k in p_augmentation)
x = data_transforms.random_lossless(x, losless_aug, rng)
return x
def data_prep_function_train(x, p_transform=p_transform, p_augmentation=p_augmentation, **kwargs):
x = x.convert('RGB')
x = np.array(x)
x = np.swapaxes(x,0,2)
x = x / 255.
x = x.astype(np.float32)
x = data_transforms.random_lossless(x, p_augmentation, rng)
return x
def data_prep_function_train(x,
p_transform=p_transform,
p_augmentation=p_augmentation,
**kwargs):
x = np.array(x, dtype=np.float32)
x = data_transforms.channel_zmuv(x,
img_stats=channel_zmuv_stats,
no_channels=4)
x = data_transforms.random_lossless(x, p_augmentation, rng)
return x
def data_prep_function_train(x, p_transform=p_transform, p_augmentation=p_augmentation, **kwargs):
x = x.convert('RGB')
x = np.array(x)
x = np.swapaxes(x,0,2)
x = x / 255.
x -= mean
x /= std
x = x.astype(np.float32)
x = data_transforms.random_lossless(x, p_augmentation, rng)
x = np.pad(x,[[0,0],[22,21],[22,21]],"edge")
return x
def data_prep_function_train(x,
p_transform=p_transform,
p_augmentation=p_augmentation,
**kwargs):
x = x.convert('RGB')
x = np.array(x)
x = np.swapaxes(x, 0, 2)
x = x / 255.
x -= mean
x /= std
x = x.astype(np.float32)
x = random_crop(x, p_augmentation['aug_out_size'][0],
p_augmentation['aug_out_size'][0], rng)
x = data_transforms.random_lossless(x, p_augmentation, rng)
random_crop
return x