ML & Animation Thesis (2019)

•

Human motion convolutional autoencoders using different rotation representations

For the culmination of my master in Computer Science at Concordia University (2019), I worked on a research thesis combining knowledge of 3D animation & Machine Learning.
The implementations in the research project were developed in C++, Python, and Keras.

Representation of autoencoder dimensions and structure for Axis Angles and Quaternions

In a nutshell the research proposes the application of four different techniques of animation storage (Axis Angle, Quaternions, Rotation Matrices and Euler Angles), in order to determine the advantages and disadvantages of each method through the training and evaluation of autoencoders for reconstructing and denoising parsed data, when passing through a convolutional neural network.

My results show that the most accurate method qualitatively is Quaternions, followed by Rotation Matrices, Euler Angles and finally with the least accurate results, Axis Angles.

Consistent denoising results were achieved in the representations, up until sequences with 25% of added gaussian noise.

The full playlist of results is available at the following link:
https://www.youtube.com/watch?v=ZNI6BoclFjA&list=PL4LdhnnKucce6L0DmSZMfOdFPAEaUbf_c

For the PDF version of the thesis it can be found at

https://github.com/vladimirdlc/UniversalMotion-Thesis/raw/master/ConcordiaThesis.pdf