Tfjs: How to train my own postnet model?
https://github.com/tensorflow/tfjs-models/tree/master/posenet
How to train my own postnet model?
lijing271
All 6 comments
@oveddan Any thoughts?
caisq
on 19 Jun 2018
@lijing271 take a look at https://medium.com/tensorflow/real-time-human-pose-estimation-in-the-browser-with-tensorflow-js-7dd0bc881cd5, it has some links to a couple of papers describing PoseNet and its setup. I believe the model in that repository was converted from one that those researchers had trained.
tafsiri
on 26 Jun 2018
@tafsiri ohh锛宼hanks very much.
lijing271
on 27 Jun 2018
@lijing271 Have you found out how to train the model?
tonyguo1987
on 17 Jul 2018
@tonyguo1987
Not yet, but I found that the single pose model is the same as openpose
lijing271
on 17 Jul 2018
So the current models that PoseNet uses are all multi-pose models. Both single pose and multi-pose estimation use the same models, they just have different algorithms to decode the outputs.
It's not the same as the OpenPose model.
All of the models that were released with PoseNet were trained based on research in the paper:
In particular, section 3.1 explains how pose estimation is trained.
oveddan
on 17 Jul 2018
Related issues
weiji14
路
3Comments
JasonShin
路
4Comments
rumschuettel
路
3Comments
neilljordan
路
4Comments
tpreusse
路
4Comments
Most helpful comment
So the current models that PoseNet uses are all multi-pose models. Both single pose and multi-pose estimation use the same models, they just have different algorithms to decode the outputs.
It's not the same as the OpenPose model.
All of the models that were released with PoseNet were trained based on research in the paper:
PersonLab: Person Pose Estimation and Instance Segmentation with a Bottom-Up, Part-Based, Geometric Embedding Model
In particular, section 3.1 explains how pose estimation is trained.