anatolix / rmpe_dataset_server

Realtime Multi-Person Pose Estimation data server. Used as a training and validation data provider in training process.

BROKEN

This fork of server is intentionally broken. Each picture augmented in same way, order of pictures always same Used for regression testing of py_rmpe_server.py Do not use for training

RMPE Dataset Server

This server generates augmented samples for the project keras_Realtime_Multi-Person_Pose_Estimation Source samples are retrieved from the hdf5 dataset file and processed in realtime using random cropping, rotations, flipping and scaling.

Prerequisites

• OpenCV
• Boost
• Hdf5
• libzmq
• cppzmq

Building

  mkdir build
  cd build
  cmake ..
  make

Usage

Usage:

./rmpe_dataset_server [DATASET] [PORT]

DATASET - path to the hdf5 dataset generated by the tool generate_hdf5.py

PORT - port number where augmented data are emitted.

Basically, you need to start a separate server for training dataset and validation dataset. The procedure of training looks something like:

1. Start training data server in the first terminal session ./rmpe_dataset_server ../../keras_Realtime_Multi-Person_Pose_Estimation/dataset/train_dataset.h5 5555
2. Start validation data server in a second terminal session ./rmpe_dataset_server ../../keras_Realtime_Multi-Person_Pose_Estimation/dataset/val_dataset.h5 5556
3. Train the model in a third terminal python train_pose.py

Troubleshooting

If you see the error

CMake Error at CMakeLists.txt:7 (find_package):
  By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH

then export the path to your OpenCV location (ex. export OpenCV_DIR=/usr/local/Cellar/opencv/3.3.0_3/share/OpenCV)