DeepPose

NOTE: This is not official implementation. Original paper is DeepPose: Human Pose Estimation via Deep Neural Networks.

Requirements

• Python 3.5.1+

• Chainer 1.13.0+

• numpy 1.9+

• scikit-image 0.11.3+

• OpenCV 3.1.0+

I strongly recommend to use Anaconda environment. This repo may be able to be used in Python 2.7 environment, but I haven't tested.

Installation of dependencies

pip install chainer
pip install numpy
pip install scikit-image
# for python3
conda install -c https://conda.binstar.org/menpo opencv3
# for python2
conda install opencv

Dataset preparation

bash datasets/download.sh
python datasets/flic_dataset.py
python datasets/lsp_dataset.py
python datasets/mpii_dataset.py

• FLIC-full dataset

• LSP Extended dataset

• MPII dataset

• Annotation

• Images

MPII Dataset

• MPII Human Pose Dataset

• training images: 18079, test images: 6908

• test images don't have any annotations

• so we split trining imges into training/test joint set

• each joint set has

• training joint set: 17928, test joint set: 1991

Start training

Starting with the prepared shells is the easiest way. If you want to run train.py with your own settings, please check the options first by python scripts/train.py --help and modify one of the following shells to customize training settings.

For FLIC Dataset

bash shells/train_flic.sh

For LSP Dataset

bash shells/train_lsp.sh

For MPII Dataset

bash shells/train_mpii.sh

GPU memory requirement

• AlexNet

• batchsize: 128 -> about 2870 MiB

• batchsize: 64 -> about 1890 MiB

• batchsize: 32 (default) -> 1374 MiB

• ResNet50

• batchsize: 32 -> 6877 MiB

Prediction

Will add some tools soon