FER2025: Explainable Binary Gender-Aware Facial Emotion Detection via Hybrid Deep Learning and Customized Visual Dataset

This dataset supports facial emotion recognition projects and includes gender classification features. The metadata maps 7,386 images containing 11,253 facial annotations sourced from Unsplash and commercial platforms. Due to strict commercial copyright terms, raw image files are withheld, and we publicly share the comprehensive metadata schema instead. The dataset features 12 distinct classes covering six primary emotions across male and female subjects, split into training (70%), validation (20%), and testing (10%) sets. The file metadata_FER2025.csv structures this data by detailing bounding boxes, class designations, and subset splits to enable efficient object detection training.