Recurrent Semi-supervised Classification and Constrained Adversarial Generation with Motion Capture Data.

RSS Source
Authors
FĂ©lix G. Harvey, Julien Roy, Christopher Pal

We explore recurrent encoder multi-decoder neural network architectures forsemi-supervised sequence classification and reconstruction. We find that theuse of multiple reconstruction modules helps models generalize in aclassification task when only a small amount of labeled data is available. Ourclassification experiments are conducted using three well known skeletal motiondatasets. We also explore a novel formulation for future predicting decodersbased on conditional recurrent generative adversarial networks. We furtherpropose both soft and hard constraints for transition generation derived fromdesired physical properties of synthesized future movements and desiredanimation goals. We find that using such constraints allow to stabilizetraining for recurrent adversarial architectures for animation generation.

Stay in the loop.

Subscribe to our newsletter for a weekly update on the latest podcast, news, events, and jobs postings.