HyperPose

HyperPose is a library for building human pose estimation systems that can efficiently operate in the wild.

News: The PoseProposal inference model is released! See the HyperPose models on Google Drive.

Features

HyperPose has two key features, which are not available in existing libraries:

Flexible training platform: HyperPose provides flexible Python APIs to build many useful pose estimation models (e.g., OpenPose and PoseProposalNetwork). HyperPose users can, for example, customize data augmentation, use parallel GPUs for training, and replace deep neural networks (e.g., changing from ResNet to MobileNet), thus building models specific to their real-world scenarios.
High-performance pose estimation: HyperPose achieves real-time pose estimation though a high-performance pose estimation engine. This engine implements numerous system optimizations: pipeline parallelism, model inference with TensorRT, CPU/GPU hybrid scheduling, and many others. This allows HyperPose to run 4x faster than OpenPose and 10x faster than TF-Pose.

Get Started

You can install HyperPose and learn its APIs through Documentation.

Example

We provide an example to show human pose estimation achieved by HyperPose. You need to install CUDA Toolkit 10+, TensorRT 7+, OpenCV 3.2+ and gFlags (cmake version), and enable C++ 17 support. Once the prerequisite are met, run the following script:

sudo apt -y install git cmake build-essential subversion libgflags-dev libopencv-dev
git clone https://github.com/tensorlayer/hyperpose.git && cd hyperpose
sh scripts/download-test-data.sh                       # Install data for examples.
sh scripts/download-tinyvgg-model.sh                   # Install tiny-vgg model.
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release && cmake --build . # Build library && examples.
./hyperpose-cli                                        # The ouput images will be in the build folder.

Performance

We compare the prediction performance of HyperPose with OpenPose 1.6 and TF-Pose. We implement the OpenPose algorithms with different configurations in HyperPose. The test-bed has Ubuntu18.04, 1070Ti GPU, Intel i7 CPU (12 logic cores). The test video is Crazy Updown Funk (YouTube). The HyperPose models (in the ONNX or Uff formats) are available here.

HyperPose Configuration	DNN Size	DNN Input Shape	HyerPose	Baseline
OpenPose (VGG)	209.3MB	656 x 368	27.32 FPS	8 FPS (OpenPose)
OpenPose (TinyVGG)	34.7 MB	384 x 256	124.925 FPS	N/A
OpenPose (MobileNet)	17.9 MB	432 x 368	84.32 FPS	8.5 FPS (TF-Pose)
OpenPose (ResNet18)	45.0 MB	432 x 368	62.52 FPS	N/A

As we can see, HyperPose is the only library that can achieve real-time human pose estimation.

License