Yoga Pose Classification Using Transfer Learning

TL;DR

This paper applies transfer learning with various deep neural networks to classify yoga poses using the Yoga-82 dataset, achieving state-of-the-art accuracy through fine-tuning and neural architecture search.

Contribution

It introduces a transfer learning approach with multiple CNN architectures and neural architecture search for improved yoga pose classification accuracy.

Findings

DenseNet-121 achieved 85% top-1 accuracy

Top-5 accuracy reached 96%

Outperformed existing state-of-the-art results

Abstract

Yoga has recently become an essential aspect of human existence for maintaining a healthy body and mind. People find it tough to devote time to the gym for workouts as their lives get more hectic and they work from home. This kind of human pose estimation is one of the notable problems as it has to deal with locating body key points or joints. Yoga-82, a benchmark dataset for large-scale yoga pose recognition with 82 classes, has challenging positions that could make precise annotations impossible. We have used VGG-16, ResNet-50, ResNet-101, and DenseNet-121 and finetuned them in different ways to get better results. We also used Neural Architecture Search to add more layers on top of this pre-trained architecture. The experimental result shows the best performance of DenseNet-121 having the top-1 accuracy of 85% and top-5 accuracy of 96% outperforming the current state-of-the-art result.