Facial Expression Image Data AFFECTNET YOLO Format

About Dataset

AffectNet is a comprehensive dataset of facial expressions containing around 0.4 million images. Since each image is manually labeled, it shows one of eight emotions: neutral, happy, angry, sad, fear, surprise, disgust, or contempt. This dataset also indicates the intensity of emotions in terms of valence (positivity or negativity) and arousal (emotional intensity). Therefore, researchers find it valuable for classifying facial expressions based on their impact on others.

The dataset consists of around 0.4 million images, each meticulously labeled to represent one of eight emotions: neutral, happy, angry, sad, fear, surprise, disgust, or contempt. Moreover, it includes indicators for the intensity of emotions, specifying valence (positivity or negativity) and arousal (emotional intensity). As a result, this dataset is invaluable for researchers and developers aiming to classify facial expressions and understand their impact on others.

The dataset, which was originally created by Noam Segal and is available on Kaggle, has undergone several modifications. To make it more accessible for computers with limited memory, the resolution of all images has been decreased to 96×96 pixels. Therefore, every image in this dataset is now exactly 96×96 pixels in size. These changes significantly enhance usability on devices with restricted memory capacity.

Dataset

This dataset, called “AFFECTNET YOLO Format,” is designed specifically for a project using YOLO for detecting facial expressions. To make it suitable for this purpose, the dataset is organized into separate folders for training, testing, and validation. Additionally, the images have been renamed and there are corresponding text files that contain annotations for each image.

In AFFECTNET, there are 8 emotion classes, i mapped them such as:

0- Anger
1- Contempt
2- Disgust
3- Fear
4- Happy
5- Neutral
6- Sad
7- Surprise