def test_tps_param__tps_scal(self, tps_scal):
from eddata.utils.tps import (
tps_parameters,
make_input_tps_param,
ThinPlateSpline,
)
tf.random.set_random_seed(42)
bs = 1
scal = 1.0
rot_scal = 0.1
off_scal = 0.1
scal_var = 0.05
augm_scal = 1.0
tps_param_dic = tps_parameters(bs, scal, tps_scal, rot_scal, off_scal,
scal_var, augm_scal)
image = get_test_image()
coord, vector = make_input_tps_param(tps_param_dic)
t_images, t_mesh = ThinPlateSpline(image, coord, vector,
image.shape[1], image.shape[-1])
with plt.rc_context({"figure.figsize": [10, 5]}):
plt.subplot(121)
plt.imshow(np.squeeze(image[0]))
plt.subplot(122)
plt.imshow(np.squeeze(t_images[0]))
return plt.gcf()
def test_tps_no_transform_params(self):
from eddata.utils.tps import (
make_input_tps_param,
ThinPlateSpline,
tps_parameters,
no_transformation_parameters,
)
tf.random.set_random_seed(42)
trf_args = no_transformation_parameters(1)
tps_param_dic = tps_parameters(**trf_args)
image = get_test_image()
coord, vector = make_input_tps_param(tps_param_dic)
t_images, t_mesh = ThinPlateSpline(image, coord, vector,
image.shape[1], image.shape[-1])
with plt.rc_context({"figure.figsize": [10, 5]}):
plt.subplot(121)
plt.imshow(np.squeeze(image[0]))
plt.subplot(122)
plt.imshow(np.squeeze(t_images[0]))
return plt.gcf()
def make_tps(self, views, tps_parameters):
"""make tps on views
Parameters
----------
views : list of tf.tensor
list of 2 image sensors shaped [N, H, W, C].
tps_parameters: dict:
tps transformation args
Returns
-------
"""
img_batch = tf.concat(views, axis=0)
bs_doubled = img_batch.shape.as_list()[0]
tps_param_dict = tps.tps_parameters(bs_doubled, **tps_parameters)
coord, vector = tps.make_input_tps_param(tps_param_dict)
t_images, t_mesh = tps.ThinPlateSpline(img_batch, coord, vector,
img_batch.shape[1],
img_batch.shape[-1])
augmented_views = tf.split(t_images, 2, axis=0)
return augmented_views