Description

Hongbo's Boxing Punch Classification Project on Roboflow

https://universe.roboflow.com/hongbo-wei/boxing-punch-classification

Project Overview

The BoxingHub Computer Vision Project aims to enhance boxing training and education by leveraging advanced AI techniques. This project focuses on developing a computer vision model capable of accurately classifying various boxing punches, such as jabs, crosses, hooks, and uppercuts, using annotated images. By providing real-time feedback to users, the project seeks to improve training effectiveness and self-assessment for boxing enthusiasts of all levels. The model is integrated into a comprehensive web platform, BoxingHub, designed to be an all-in-one resource for boxing knowledge and training.

Descriptions of Each Class Type

6 classes in total

Jab: A quick, straight punch thrown with the lead hand. It is often used to measure distance, set up combinations, and keep the opponent at bay.
Cross: A powerful, straight punch delivered with the rear hand. It typically follows a jab and is used to capitalize on openings created by the lead hand.
Hook: A semi-circular punch thrown with the lead hand, targeting the side of the opponent's head or body. It is effective at close range and often used in combinations.
Uppercut: A vertical punch directed upwards with either hand, aiming for the opponent's chin or body. It is particularly effective against opponents who lean forward or have a low guard.
No Punch:
Bag:

Current Status and Timeline

Current Status: The project is in the active development phase, focusing on fine-tuning the computer vision model and integrating it into the BoxingHub platform. Initial model training has been completed, and the model is undergoing iterative improvements based on user feedback and additional data collection.
Timeline:
- Month 1-2: Data collection and annotation, initial model training.
- Month 3: Model fine-tuning and validation.
- Month 4: Integration with the BoxingHub platform and user testing.
- Month 5: Refinement based on feedback and performance evaluation.
- Month 6: Final deployment and public release.

Links to External Resources

BoxingHub Website: Access the platform to explore boxing training resources and test the computer vision model.
Roboflow Documentation: Comprehensive guide on how to use Roboflow for computer vision projects.
Annotated Dataset: A link to the publicly available boxing punch dataset used for model training and evaluation.

Contribution and Labeling Guidelines

We welcome contributions from the community to enhance the dataset and model accuracy. Here are some guidelines to ensure consistent data quality:

Data Quality: Submit high-resolution images that clearly depict the desired punch type. Avoid blurry or low-contrast images that may confuse the model.
Annotation Accuracy: When labeling images, ensure that the bounding boxes accurately encompass the punching action without including excessive background or irrelevant objects.
Class Distribution: To maintain balanced model performance, contribute a diverse set of images for each class type (jab, cross, hook, uppercut).
Ethical Considerations: Ensure all contributed images comply with privacy and copyright regulations. Only submit images you have the right to share and use for training purposes.

By following these guidelines, contributors can help improve the BoxingHub Computer Vision Project, making boxing training more accessible, effective, and data-driven.