hmxf / DeepPoseKit

A Toolkit for Pose Estimation using Deep Learning, with some bug fixes.

DeepPoseKit

There are some little fixes for the Original Repo, reference the Original Document for more details.

Here are my install steps which have been tested on the following platforms:

• Ubuntu 20.04.5 LTS ARM on VMWare Fusion hosted on Apple Macbook with M Series CPU

• Ubuntu 20.04.6 LTS on Microsoft WSL2 hosted on Generic x86_64 PC

You can start from any stage, but start from any small steps are not recommended, unless you are aware what you are doing.

---

Install Guide

• Configure your system

  1. Update system and packages

     sudo apt update && sudo apt update && sudo apt upgrade -y && sudo apt autoremove -y && sudo apt update && sudo apt upgrade
     

  2. Install tools

     sudo apt install nano git zsh tree htop curl wget screen tmux openssh-server net-tools gcc make cmake 
     

• Install conda Environment

  1. Download Miniconda

     wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-`uname -m`.sh
     

  2. Install Miniconda

     chmod +x Miniconda3-latest-Linux-`uname -m`.sh && ./Miniconda3-latest-Linux-`uname -m`.sh
     

• Create Work Environment

  1. Create Virtual Environment

     conda create -n tf tensorflow
     

  2. Enter Virtual Environment

     conda activate tf
     

  3. Install dependencies by installing and removing pre-built version of deepposekit and scikit-learn with pip

     pip install deepposekit scikit-learn keras-core
     
     pip uninstall deepposekit scikit-learn
     

• Build deepposekit and scikit-learn from source to avoid some errors mostly caused by Architecture diffirences of your CPU

  1. Fetch source of scikit-learn from GitHub

     git clone https://github.com/scikit-learn/scikit-learn
     

  2. Install dependencies needed by compile progress

     pip install cython wheel numpy scipy
     

  3. Compile and Install scikit-learn

     cd scikit-learn
     
     pip install -v --no-use-pep517 --no-build-isolation -e .
     

  4. Test Install and following command should execute without error

     python -c "import sklearn; sklearn.show_versions()"
     

• Install DeepPoseKit

  1. Fetch source of DeepPoseKit from GitHub

     cd ~ 
     
     git clone https://github.com/hmxf/DeepPoseKit
     

  2. Install DeepPoseKit

     cd DeepPoseKit
     
     python setup.py develop
     

  3. Setup architecture-related Environment Variable

     ./scripts/setup.sh
     

• Test your Installation with pre-downloaded data within 5 miniutes ;)

  Pre-downloaded data is located in data/ directory and has been used by scripts/train.py and scripts/predict.py for fast test purpose only.

  python scripts/train.py
  

  After a model train process, you can verify if model data has been generated successfully under data/ directory. If your model data has been stored as a single file data/saved_model.h5, then you can view its structure and weight parameters by using scripts/hdf5_file_reader.py script.

  python scripts/hdf5_file_reader.py
  

  One more step, you can use the trained model to do some predictions.

  python scripts/predict.py