3d-pose-baseline-keras

Keras implementation of 3d-pose-baseline.

from WiderFaceDataset

Demo

Download Model

python download_model.py

Predict

Predict using Caffe OpenPose and Keras

python predict.py

How to Train

Create Dataset

Dump training data from 3d-pose-baseline using export_dataset.py

https://github.com/una-dinosauria/3d-pose-baseline

Check Exported Data

python plot.py

Train

Training using Keras

python train.py

This is a pretrained output

http://www.abars.biz/keras/3d-pose-baseline.hdf5

http://www.abars.biz/keras/3d-pose-baseline-mean.h5

Convert to Caffemodel

python convert.py

http://www.abars.biz/keras/3d-pose-baseline.caffemodel

http://www.abars.biz/keras/3d-pose-baseline.prototxt

About 3d-pose-baseline

Architecture

3d-pose-baseline predict 3d pose from 2d pose.

Input is 16 keypoint. Each keypoint has 2 axis.

Output is 16 keypoint. Eash keypoint has 3 axis.

Output should be denormalize using mean value.

Mean value has 32 keypoint, So you should remove unused dimension. Mean value is sparse.

Original work

https://github.com/una-dinosauria/3d-pose-baseline

@inproceedings{martinez_2017_3dbaseline, title={A simple yet effective baseline for 3d human pose estimation}, author={Martinez, Julieta and Hossain, Rayat and Romero, Javier and Little, James J.}, booktitle={ICCV}, year={2017} }

Related work

OpenPose to 3dpose

https://github.com/miu200521358/3d-pose-baseline-vmd/blob/master/src/openpose_3dpose_sandbox_vmd.py