This is the code for an attempt to plug the code from
Julieta Martinez, Rayat Hossain, Javier Romero, James J. Little.
A simple yet effective baseline for 3d human pose estimation.
In ICCV, 2017. https://arxiv.org/pdf/1705.03098.pdf.
into Android.
The code in this repository was mostly written by Julieta Martinez, Rayat Hossain Javier Romero and Arash Hosseini.
I have modified it to use TfLite Posenet as the 2d keypoint model, and to be runnable in an Android app. The 2d->3d model used is one pretrained on SH detections, downloadable here.
From the paper by Martinez et al.: "We provide a strong baseline for 3d human pose estimation that also sheds light on the challenges of current approaches. Our model is lightweight and we strive to make our code transparent, compact, and easy-to-understand."
Here's a preview of the model running in real time, using TfLite Posenet:
- h5py
- tensorflow 1.0 or later
- You need access to the Human3.6M dataset. If you don't have access, you can't run this code.
- Clone this repository and get the data.
- If you got this far, wait for further instructions. The code is under construction.
You can test the pre-trained model by downloading it at the link above, decompressing the file at the top level of this project, and calling
python src/predict_3dpose.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --sample --load 4874200