This project aims to be a benchmark on classifying yoga poses using pure Convolutional Neural Networks (CNN) and without using any specialized technique. This is achieved by using a novel idea of using ConvNext as a backbon to extract metafeatures from images and using conventional classifying machine learning techniques to correcty predict yoga poses with higher accuracy.
📦 HYPCNet
├─ train.py
utils.py
├─ yoga82
│ ├─ yoga_train
│ │ ├─ class_6
│ │ ├─ class_20
│ │ └─ class_82
│ └─ yoga_test
│ ├─ class_6
│ ├─ class_20
│ └─ class_82
└─ out
├─ models
├─ test_{class_name}.csv
├─ {model}_{class_name}_training_metrics.csv
└─ models
└─ {model}_{class_name}_new_best_model.pth
©generated by Project Tree Generator
- Hybrid Model Architecture: Integration of ConvNeXt with traditional ML models like XGboost, RandomForest etc.
- Few Shot Like Learning Abilities: Metafeatures extraction helps in classifying classes with limitations.
- Extensive Assesment: Detailed metrics comparisions with other STOA models and contemporary models.
To download the dataset: https://forms.gle/tzVHwzbzCEYzZd9W8
More details about the dataset: https://sites.google.com/view/yoga-82/home
Kindly give proper citation to the original authors
We would like to thank the authors of the Yoga-82 repository for providing a solid foundation for our work. Their initial framework was essential in developing our enhanced model.
This project is licensed under the MIT License.