heytherewill / yoga-82

Yoga-82 dataset evaluation

Yoga-82 dataset evaluation keras implmentation

Dataset preparation

• Vist this link and fill out the form to download the dataset
• Download images using the image_fetcher.ps1.
• Manually clean up the images folder by removing wrongly downloaded ones.
• Run clean_up_sets.ps1 to update the contents of the train/test splits.
• Use train and test splits provided in the dataset.

Usage

Use pip to install keras, keras_applications, sklearn, Pillow and tensorflow. Check models.py for different existing models and modified hierarchical models. And modify train_yoga.py accordingly.

Train model python train_yoga.py

Reference

Verma, M., Kumawat, S., Nakashima, Y., & Raman, S. (2020). Yoga-82: a new dataset for fine-grained classification of human poses. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (pp. 1038-1039).