A Cross-Temporal Individual Identification Dataset for Large Yellow Croaker (Larimichthys crocea)

A Cross-Temporal Individual Identification Dataset for Large Yellow Croaker (Larimichthys crocea) This repository contains the image dataset and corresponding biometric metadata for the individual recognition of Larimichthys crocea. The data spans a 7-week interval (from 8 weeks to 1 week before spawning) to evaluate cross-temporal identification capabilities. ⚠️ Current Release Note: To facilitate immediate reproducibility and demonstrate the data structure for peer review, we are initially releasing a curated subset of 100 individuals. The complete large-scale dataset is undergoing final alignment and will be incrementally updated. 📂 Dataset Structure 1. Classification Dataset (Classification/) Content: Cropped images of the main fish body for metric learning/classification. Structure: Each folder represents a unique individual. Inside each folder, it contains bilateral images (both left and right sides) captured at two time points: 8 weeks before spawning and 1 week before spawning. 2. Detection Dataset (Detection/) Content: Standard object detection dataset format for training YOLO-like models to crop the fish bodies. Structure: Contains standard subdirectories: JPEGImages (raw images), Annotations (bounding box labels), and ImageSets (train/val splits). 3. Metadata & Ground Truth This dataset provides rigorous ground truth by pairing visual tags with internal PIT (Passive Integrated Transponder) tags: 8 weeks before spawning.xlsx: Biometric data for the initial enrollment (8 weeks prior). Includes visual tag info, internal PIT numbers, body weight, body length, and body height. 1 weeks before spawning.xlsx: Biometric data for the second capture (1 week prior). Includes visual tag info, PIT numbers, body weight, and breeding pool IDs. match.xlsx: The absolute mapping relationship between the external visual tag numbers and the internal PIT tag IDs. This acts as the ultimate ground truth for model evaluation.