CS249r Final Project: The Saga of DeepMimic, Dance Dance Robot, and Quadraped Pose Estimation

This blog post is under construction, pending getting DeepPoseCut to train. For now, please see the posts

http://orangenarwhals.com/hexblog/2019/12/06/CS249r-further-research/

http://orangenarwhals.com/hexblog/2019/12/04/249r/

These are on my secondary blog which consists of me taking notes for myself, essentially, and are not cleaned up, but reflect some of the effort put into this project. They reflect the failed attempts to set up DeepMimic and in particular Motion_Reconstruction from the Skills From Video paper, which relies on several third-party libraries in horrible recursive/nesting structure of poorly documented repositories.

Not super proud of the work, but here’s the poster summarizing it. Just didn’t have time…

Full size Poster:

https://github.com/nouyang/cs249r_finalproject/blob/master/poster.png

Stereo

Annotation

  • “Using DeepLabCut for 3D markerless pose … – bioRxiv.” 24 Nov. 2018, https://www.biorxiv.org/content/10.1101/476531v1. Accessed 9 Dec. 2019.

Closeup of video, dog 1

Dog 2, checkerboard

 

BVH Player / mocap quadraped exists! Loaded data from AI4Animation 2018 SIGGRAPH paper

 

 

Links

  • “AI4Animation/ReadMe.txt at master · sebastianstarke … – GitHub.” https://github.com/sebastianstarke/AI4Animation/blob/master/AI4Animation/Assets/Demo/SIGGRAPH_2018/ReadMe.txt. Accessed 9 Dec. 2019.
  • “BVH player.” http://lo-th.github.io/olympe/BVH_player.html. Accessed 9 Dec. 2019.