An implementation of quaternions for and written in tensorflow. Fully differentiable. Licensed under Apache 2.0 License.
Note: This project is currently in alpha status. Some functions have not even been tested yet.
The tfquaternion module provides an implementation of quaternions as a tensorflow graph.
The quaternion value can either be represented as tf.Tensor or tf.Variable.
As all operations are derivable, the module can be used to optimize a rotation of
points in 3D space, given that a tf.Variable is used to represent the value.
Other awesome features are:
- Operations are scoped, so they appear nice and clean in your tensorboard graph.
- Operators are implemented.
Let's take a look at a simple rotation:
>>> import tfquaternion as tfq
>>> import tensorflow as tf
>>> s = tf.Session()
>>> points = tf.constant([[1, 0, 0], [0, 1, 0], [0, 0, 1]], dtype=tf.float32)
>>> quat = tfq.Quaternion([0, 1, 0, 0]) rotate by 180 degrees around x axis
>>> s.run(tf.matmul(quat.as_rotation_matrix(), points))
array([[ 1., 0., 0.],
[ 0., -1., 0.],
[ 0., 0., -1.]], dtype=float32)
If you'd like to have a certain feature please check the ToDo file first before opening an issue.