PocketPose

An open-source framework designed for real-time pose estimation on consumer devices, such as smartphones, tablets, and laptops.

PocketPose provides a unified interface for performing inference on a range of pose estimation models, and includes tools for converting and optimizing models for mobile deployment.

PocketPose
Unified Inference API logo

Unified Inference API

A single API for all your inference needs across frameworks and devices.
Conversion Tools logo

Conversion Tools

Convert your existing PyTorch models to ONNX and TFLite for deployment.
Deployment Engine logo

Deployment Engine

We provide Python and Kotlin APIs for deploying your models to laptops and mobile devices.

Citation

If you find PocketPose useful in your research, please consider citing the following paper:

@misc{khan2023pocketpose,
      title={PocketPose: Real-Time Human Pose Estimation on Mobile Devices - A Comprehensive Study}, 
      author={Muhammad Saif Ullah Khan},
      year={2023},
      journal={arXiv preprint arXiv:2103.16345},
      eprint={2103.16345},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}