Leveraging Pre-trained ResNet-18 with Transfer Learning for Yoga Posture Classification

This research paper explores the application of pre-trained ResNet-18 with transfer learning for the classification of yoga postures. The study utilizes a dataset comprising images of various yoga poses taken from Kaggle. Through fine-tuning, the pre-trained model achieved impressive training accuracy of 9 8 %. Furthermore, on unseen data, the model maintained accuracy of 9 4. 9 3 %. Advantages and disadvantages of the methodology are discussed, along with experimental setup details and metrics evaluation. The findings underscore the efficacy of leveraging pre-trained models and transfer learning techniques for complex image classification tasks in specialized domains like yoga postures. However, a notable limitation arises in the difficulty of removing noise during model implementation, leading to a decrease in confidence scores. For future work, the project aims to address this limitation by focusing on noise removal techniques while maintaining accuracy and confidence scores, thus enhancing the robustness of the classification model.