Facial Expressions Detections Model Using YOLOv5 Trained with Custom Images

Facial expressions in disciplines including psychology, human-computer interaction, and computer vision. In this study, we introduce a unique method for detecting facial emotions utilizing the YOLOv5 object detection framework. We created a unique collection of face expression photographs using our own camera to train our model. The makesense.ai web tool, which offered effective annotation capabilities for precisely recognizing and classifying facial expressions, was subsequently used to label these photographs. To provide a wide variety of training samples, the dataset included a variety of expressions, including neutral, surprise, rage, happiness, and sadness. We carried out a thorough ablation study to examine the effects of various variables on the effectiveness of the facial expressions detecting model. To comprehend their impact on the model's accuracy and robustness, we investigated the effects of dataset size, training length, hyper-parameter tuning, and architectural modifications.