pocketpose 1.0.0a2.post0

An inference library for real-time human pose estimation.

PocketPose Python API

Developed by Saif Khan

 

📘Documentation | 🛠️Installation | 👀Model Zoo | 📜Papers | 🆕Update News | 🤔Reporting Issues |

Introduction

PocketPose is an open-source library that facilitates fast and efficient pose estimation on mobile and edge devices. It supports TensorFlow Lite (TFLite) models and provides a collection of pre-trained models ready to use out of the box.

Features

• High-performance pose estimation on mobile devices
• Support for TensorFlow Lite models
• Collection of pre-trained models for various use-cases
• Easy-to-use API for custom model deployment

Installation

PocketPose can be installed from source:

pip install pocketpose

This installs the CPU-only version of PocketPose. To install the GPU version, use:

pip install pocketpose-gpu

Please refer to the installation guide for detailed installation instructions.

Quick Start

Running pose estimation on an image is as simple as:

import pocketpose as pp

# Initialize the image inferencer
inferencer = pp.ImageInferencer(
  pose_model="pose-model-alias",
  det_model="detection-model-alias",
  device="cpu",
)

# Estimate keypoints from the image
image_path = "path/to/image.jpg"  # JPEG or PNG
keypoints = inferencer.infer(image_path)

# Visualize and save the keypoints
inferencer.visualize(image_path, keypoints, save_path="path/to/save.jpg")

# Alternatively, visualize directly during inference
# inferencer.infer(image_path, visualize=True, save_path="path/to/save.jpg")

# If save_path is omitted, visualization is displayed on screen

For more detailed usage and a list of all available models, check out our usage guide.

License

PocketPose is released under the CC BY-NC 4.0 license. See the LICENSE for more details.