0

I have been reading this paper and there is point not clear to me. For their algorithm they are using body parts, but i couldn't find how they obtain the body part. Is there any straight forward method for obtaining body parts?

paper link

seleucia
  • 1,046
  • 4
  • 17
  • 30

1 Answers1

1

In the supplementary material published by Microsoft at this link, authors explain they used labeled 3D models to train the algorithm. So the body parts used for the training were "hand-labeled" for each of the models used.

This image should help to figure out how the labeling process were made:

3D labeled models

Vito Gentile
  • 13,336
  • 9
  • 61
  • 96
  • Hi Vito thank you answer, But they were mentioning they used real data also for building training trees. How can they partition the body on real depth image – seleucia Mar 03 '16 at 14:14
  • I read authors mention some "mocap data", so I think they probably used [mo-cap](https://en.wikipedia.org/wiki/Motion_capture) to move such 3D models and morph them in several poses. – Vito Gentile Mar 03 '16 at 14:50