Recurrent semi-supervised classification and constrained adversarial generation with motion capture data
作者：
Highlights：
• A new semi-supervised architecture of the recurrent encoder, multi-decoder type
• A new realistic partitioning of a common motion capture data set (HDM05)
• Empirical evidence of the generalization properties of our method on such a partitioning
• A new adversarial recurrent transition generator for continuous trajectories
• Data-driven soft constraints stabilizing training for recurrent adversarial networks
摘要
•A new semi-supervised architecture of the recurrent encoder, multi-decoder type•A new realistic partitioning of a common motion capture data set (HDM05)•Empirical evidence of the generalization properties of our method on such a partitioning•A new adversarial recurrent transition generator for continuous trajectories•Data-driven soft constraints stabilizing training for recurrent adversarial networks
论文关键词：Action recognition,Motion capture,Semi-supervised learning,Recurrent neural networks,Generative adversarial networks,Transition generation
论文评审过程：Received 12 June 2017, Revised 10 July 2018, Accepted 10 July 2018, Available online 20 July 2018, Version of Record 15 August 2018.
论文官网地址：https://doi.org/10.1016/j.imavis.2018.07.001
