Multistage Fish Eye Dataset

This dataset, titled the Multistage Fish Eye Dataset (MFED), is specifically designed to facilitate research in fine-grained visual classification for food safety and quality control. It contains a comprehensive collection of 4,800 high-resolution images of fish eyes, meticulously categorized into three distinct stages of freshness: Highly Fresh, Fresh, and Not Fresh.

The biological degradation of fish is heavily reflected in the ocular features, transitioning from high-frequency specular reflections and sharp pupil edges (highly fresh) to low-frequency, cloudy, and opaque corneas (not fresh). This dataset captures these subtle, multistage morphological changes, providing an ideal benchmark for training and evaluating deep learning architectures, particularly lightweight Convolutional Neural Networks (CNNs) and customized attention mechanisms (e.g., Focal Attention, Spatial-Channel Attention) intended for edge-device deployment.