DensePoseCompare: A Comparative Study of DenseNet Models in Yoga Pose Classification

This work introduces DensePoseCompare, representing an extensive investigation of DenseNet models for precise and effective yoga position detection. A comparative study is conducted utilizing three DenseNet architectures—DenseNet121, DenseNet169, and DenseNet201—on a collection of 1551 yogic posture images, including downdog, goddess, plank, tree, and warrior2. The training dataset, which contains 1081 images, allows these models to be fine-tuned, while the testing dataset, which has 470 images, evaluates their performance. DenseNet201 outperforms other models with an outstanding 99.07% testing accuracy. The suggested method, presented in the DensePoseCompare framework, demonstrates the power of DenseNet models in the field of yoga posture classification. This comparative investigation reveals the efficiency of DenseNet architectures in capturing detailed aspects of yoga postures. The proposed model also uses the explainable AI (XAI) method LIME to give insights into how decisions are made.