Implementation of Thin Plate Spline. (See numpy implementation with thin-plate-spline library)

$ pip install torch-tps

Not yet available

import torch
from tps import ThinPlateSpline

# Some data
X_c = torch.normal(0, 1, (800, 3))
X_t = torch.normal(0, 2, (800, 2))
X = torch.normal(0, 1, (300, 3))

# Create the tps object
tps = ThinPlateSpline(alpha=0.0)  # 0 Regularization

# Fit the control and target points
tps.fit(X_c, X_t)

# Transform new points
Y = tps.transform(X)

We provide different examples in the example folder. (From interpolation, to multidimensional cases and image warping).

The elastic deformation of TPS can be used for image warping. Here is an example of tps to increase/decrease the size of the center of the image or using random control points:

Have a look at example/image_warping.py.

$ python -m build
$ python -m twine upload dist/*