This repository is for implementing YOLOv5 for hand gesture recognition. The dataset consists of 200 images of four hand gesture classes: Fist, OpenPalm, PeaceSign and ThumbsUp. The dataset was created by myself.
First, I used the Roboflow website to label my dataset and perform data augmentation, resulting in a total of 408 images.
The training dataset consisted of 368 images, while the validation and test datasets each comprised 20 images. Using this site, I imported my dataset into Google Colab. Then, following the guide provided by the YOLOv5 repository, I proceeded with the training.
During training, the model achieved a precision of 93% for all classes, with individual class precisions as follows: 92% for Fist, 91% for OpenPalm, 90% for PeaceSign, and an impressive 99% for ThumbsUp.
After training, I evaluated the model using the test dataset. Below is an example output image generated by the model:
