FER2013

FER2013, also known as the Facial Expression Recognition 2013 dataset, is a widely used benchmark dataset for the task of automatic facial expression recognition. It was released in 2013 and has since become a standard resource for researchers in the field. The dataset comprises over 35,000 facial images, each labeled with one of seven basic emotion categories: angry, disgust, fear, happy, sad, surprise, and neutral. The images were collected from the internet, and consequently, they exhibit a variety of real-world complexities such as variations in pose, illumination, occlusion, and image quality.

The FER2013 dataset is divided into three sets: training, public testing, and private testing. The training set contains the majority of the images and is used to train machine learning models. The public test set is available for researchers to evaluate their models and compare their performance against others. The private test set is kept hidden, and its results are used for the final ranking in competitions. The diversity and challenging nature of FER2013 have made it instrumental in advancing the state-of-the-art in facial expression recognition, driving the development of more robust and accurate algorithms. Despite its widespread use, challenges remain in achieving human-level accuracy on this dataset due to its inherent noise and ambiguity.