Realtime-3D-pose-for-singleperson

2D Human Pose Estimation

First run the 2d pose estimator model for generating the 2D predictions

Dependencies:

•	python3

•	tensorflow 1.4.1+
   
•	opencv3, protobuf, python3-tk

git clone https://www.github.com/ildoonet/tf-openpose

cd tf-openpose

pip install -r requirements.txt

Realtime

python run_webcam.py --model=mobilenet_thin --resize=432x368 --camera=0 --output_json /path/to/directory

3D Human Pose Estimation

Dependencies:

•	H5py

•	Tensorflow 1.0 or later
    
•	Python 3

git clone https://github.com/Uday038/Realtime-3D-pose-for-singleperson.git

cd Realtime-3D-pose-for-singleperson

mkdir data

cd data

download human3.6M data from https://drive.google.com/drive/folders/1HBGmdk9UyeOXKgqnt82GiP43SDIWcHc- and store in data folder

cd ..

Training

python train.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm –evaluateActionWise --use_2d

Generating 3D predictions from 2D predictions

python pose3D_normal.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory

Realtime

Python pose3D_realtime.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory

Prerequisites

In order to run the model in realtime, first run the 2D pose estimator followed by 3D pose estimation model.