mehmet-3emin/fer2013-cleaned

FER2013 Cleaned and Extended – mehmet-3emin

This dataset is a cleaned and extended version of the original FER2013 dataset, adapted for facial emotion recognition tasks. It was prepared as part of a senior thesis project at Mersin University in 2025, titled "Videodan Duygusallık Analizi" (Emotion Analysis from Video).

The dataset is formatted for folder-based image classification tasks and includes 7 emotion classes.

🧹 Cleaning Process

The original FER2013 dataset was reviewed and processed using the following steps:

• Removed low-resolution or blurry images (below 48x48 or heavily pixelated)
• Standardized image dimensions and file naming conventions
• Corrected mislabeled samples where clearly identifiable

🧩 Extension Details

To improve generalization and diversity:

• Additional facial images were extracted from various public video sources using OpenCV
• Individual frames containing clearly visible faces were manually cropped
• These new images were manually labeled by human annotators into 7 emotion categories: angry, disgust, fear, happy, neutral, sad, surprise

This extension allowed the dataset to include more varied lighting, angle, and expression styles, making it more robust for real-world facial emotion detection tasks.

🏷️ Class Distribution

The dataset is organized into folders by emotion class:

fer2013-cleaned/ ├── train/ │ ├── angry/ │ ├── disgust/ │ ├── fear/ │ ├── happy/ │ ├── neutral/ │ ├── sad/ │ └── surprise/ ├── test/ │ └── …

Each folder contains .jpg or .png images with RGB channels and resized to 48x48 or 224x224 resolution.

📦 Usage

from datasets import load_dataset
dataset = load_dataset("mehmet-3emin/fer2013-cleaned")