LaurenGurgiolo/9_Facial_Expressions

Facial Expressions for YOLO & ViT – 9 Emotions Dataset

Dataset ID: LaurenGurgiolo/9_Facial_Expressions Task: Facial Emotion Recognition Domain: Computer Vision, Deep Learning, Affective Computing Languages: N/A License: Refer to original Roboflow Universe sources

Dataset Description

The Facial Expressions for YOLO & ViT – 9 Emotions Dataset is a curated facial emotion recognition dataset derived from the Facial Expressions for YOLO Facial Expression Recognition – 9 Emotions Dataset, originally compiled from multiple high-quality repositories hosted on Roboflow Universe (Rimi, 2025).

This dataset has been reorganized and refined to support Vision Transformer (ViT)–based models, in addition to convolutional and YOLO-style pipelines. Specific emotional categories were selectively extracted, grouped, and restructured to ensure compatibility with ViT architectures and image classification workflows.

The dataset is intended to support research and experimentation in facial expression analysis across modern deep learning frameworks.

Emotion Classes

The dataset contains images labeled across nine emotional categories:

Angry

Disgust

Fear

Happy

Sad

Surprise

Neutral

Contempt

Calm

(Exact class naming may vary slightly depending on preprocessing and folder structure.)

Dataset Sources

The original dataset aggregates facial expression images from multiple curated repositories available through Roboflow Universe, selected for image quality, facial visibility, and emotional clarity.

Primary Source:

Facial Expressions for YOLO Facial Expression Recognition – 9 Emotions Dataset

Roboflow Universe

Curated by Rimi (2025)

Data Processing & Curation

To prepare the dataset for use with Vision Transformer (ViT) models, the following steps were performed:

Extraction of images corresponding to selected emotional categories

Consolidation of images from multiple Roboflow sources into unified class groups

Reorganization of the dataset into a classification-friendly structure

Standardization of directory layout for compatibility with Hugging Face datasets and ViT pipelines

This restructuring ensures that the dataset can be easily used for image classification tasks without requiring bounding-box annotations.

Intended Use

This dataset is suitable for:

Facial emotion recognition

Vision Transformer (ViT) model training and evaluation

CNN-based emotion classification

Transfer learning experiments

Comparative studies between YOLO-based and classification-based emotion recognition pipelines

Limitations

Images originate from multiple public sources and may vary in lighting, pose, age, ethnicity, and cultural expression.

Some emotions may be visually ambiguous or overlapping.

Class imbalance may exist depending on the original source distributions.

Ethical Considerations

Faces are sourced from publicly available datasets.

The dataset may contain demographic biases inherent to the original repositories.

Users should ensure responsible and ethical use, particularly in sensitive applications.

Citations

If you use this dataset, please cite:

Rimi. (2025). Facial Expressions for YOLO Facial Expression Recognition – 9 Emotions Dataset. Roboflow Universe.

Dataset Access from datasets import load_dataset

dataset = load_dataset("LaurenGurgiolo/9_Facial_Expressions")