This is not official implementation. Original paper is DeepPose: Human Pose Estimation via Deep Neural Networks. And the implementation model performs 3D pose esimation.

• Python 2.7.6
• Chainer 1.19.0
• NumPy 1.11.3
• six 1.10.0
• Pillow 3.4.1
• SciPy 0.17.1
• OpenCV 2.4.8
• pyquaternion 0.9.0
• paramiko 1.16.0
• scp 0.10.2
• matplotlib 1.5.1
• PyQt (QtCore, QtGui, QtOpenGL) 4.10.4
• PyOpenGL 3.1.0

First, download HumanEva Dataset to 'DeepPose/orig_data'. And execute the folowing script.

python datasets/human_eva.py
python datasets/compute_mean.py

human_eva.py performs to calculate bounding boxes of human, resize images, and modify 2D/3D poses for your Neural Networks training. compute_mean.py performs to compute a mean image of the datasets.

Just run:

python train/train_pose_net.py

If you want to run train_pose_net.py with your own settings, please check the options first by python train/train_pose_net.py --help and give customize training settings.

Just run:

python visualize/visualize_log.py

If you run train_pose_net.py on some clouds, please specify the ip address with python visialize/visualize_log.py --locate '[email protected]'. And you can see the other settings with python visialize/visualize_log.py --help.

Execute the following scripts, GUI tool will start.

python demo/stream_player.py