CleanFER25_RAF_CK

The Clean FER25-RAF-CK+ dataset is a high-quality, cleaned, and integrated facial emotion recognition (FER) dataset constructed from three widely used benchmarks: FER2013 (subset FER25), RAF-DB, and CK+. The dataset was curated to address inconsistencies, noise, and class imbalance commonly found in existing FER datasets. All images were manually reviewed and filtered to remove mislabeled, low-quality, blurred, occluded, or ambiguous samples. The final dataset contains standardized, frontal-face images categorized into seven universal emotion classes: Angry, Disgust, Fear, Happy, Neutral, Sad, and Surprise.The dataset is organized into train, validation, and test subsets, providing a balanced benchmark suitable for deep learning models, including CNNs, vision transformers, hybrid FaceNet-CBAM architectures, and emotion recognition pipelines requiring strong generalization. This dataset was developed as part of a doctoral research project on real-time facial emotion recognition for assistive smart glasses for visually impaired individuals.Clean FER25-RAF-CK+ offers researchers a reliable, noise-reduced dataset with consistent labeling and directory structure, enabling reproducible training, benchmarking, and testing of state-of-the-art FER models.