abs711/Optimizing-representations-for-gait-generation

Using encoder-decoder neural networks to learn representations of personal walking style, and generating person-specific gait for desired activities.

Optimizing-representations-for-gait-generation

Code coming soon! Details of this project can be found on the following link: https://ieeexplore.ieee.org/document/10121330

How would this work in practice

Network Architectures

We compare 4 methods: Baseline (adapted from https://arxiv.org/pdf/1511.05644.pdf), Output Regularization, Latent Regularization, and Combined Regularization.

Baseline

Output Regularization

Latent Regularization

Learned representations of personal style from the data

Different subjects are shown in different colors.

Generated Trajectories

Trajectories generated by different methods overlayed with the "Ground Truth" of the unseen target user (shown in black).