Skip to content

Latest commit

 

History

History
122 lines (95 loc) · 4.47 KB

README.md

File metadata and controls

122 lines (95 loc) · 4.47 KB


Efficient YoloPose

Efficient YoloPose is a deep learning-based architecture designed to balance computational efficiency and accuracy for real-time multi-person pose estimation. This repository contains the implementation of Efficient YoloPose, which improves upon YOLOv8-Pose by incorporating advanced lightweight techniques, reducing computational demands while maintaining high performance, particularly for resource-constrained devices.

Overview

Human pose estimation (HPE) is crucial for various real-time applications, but achieving high accuracy with low computational overhead is a challenge, especially on resource-constrained devices. Efficient YoloPose addresses this challenge by optimizing YOLOv8-Pose with lightweight components to achieve both efficiency and accuracy. Efficient design: Utilizes Depthwise Convolution, Ghost Convolution, and C3Ghost to reduce model size and complexity. Advanced attention mechanism: Integrates Squeeze Excitation (SE) attention for improved feature extraction. Superior performance: Reduces inference time, computational complexity, and parameter count compared to traditional models like YOLOv8-Pose.

Key Features

- Improved Inference Time: The proposed model reduces the inference time from 1.1 ms to 0.9 ms.
- Reduced Computational Complexity: The computational complexity decreases from 9.2 GFlops to 4.8 GFlops.
- Lower Parameter Count: The parameter count is reduced from 3.3 million to 1.3 million.
- Accuracy: The model achieves an average precision (AP) score of 78.8 on the COCO dataset, outperforming other models in both efficiency and accuracy.
- Lightweight Design: Optimized for resource-constrained devices while maintaining high performance.

Weixin Image_20241205175638

Weixin Image_20241205175712

Performance Comparison

When compared to YOLOv8-Pose, Efficient YoloPose: .
- Reduces inference time
- Lowers computational complexity
- Reduces parameter count
- Maintains competitive accuracy

Datasets Used

The model was evaluated on the following datasets: .
- COCO: Common Objects in Context, a large-scale object detection dataset.
- OCHuman: A dataset specifically designed for human pose estimation.

Installation

To use Efficient YoloPose, clone this repository to your local machine:

git clone https://github.com/malareeqi/Efficient-YoloPose.git
cd Efficient-YoloPose

Install the required dependencies:

pip install -r requirements.txt

Usage

To train this model, simply execute the following Python script: bash

%cd /Efficient-YoloPose/ultralytics/models/yolo/pose/
python train.py

Make sure you have writting the correct path before running the model.

Note

This repository is actively being improved, with continuous updates and optimizations underway. Check back for the latest improvements and updates to the model's performance and capabilities.

Citation

If you use this model in your research or projects, please cite the following:

@article{EfficientYoloPose,
    title={Resource-Aware Strategies for Real-Time Multi-Person Pose Estimation},
    author={Mohammed A. Esmail, Jinlei Wang,*,Yihao Wang, Li Sun, Guoliang Zhu, and Guohe Zhang*}
    journal={XXXX.XXXX},
    year={2024}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.