SPDD-3K: Fixed 80/10/10 Split Manifests and Annotation Audit for Solar Panel RGB Defect Detection

Fixed train/validation/test image-ID manifests (2,400/300/300), annotation audit log, dataset statistics, and co-occurrence priors for the SPDD-3K benchmark derived from the SIL-IoT Mendeley v1 solar panel anomaly dataset (DOI: 10.17632/8x7wjsr94v.1). This record does not include RGB images or bounding-box label files. Download the parent dataset from Mendeley Data and apply the manifests in splits/ to reproduce the exact 80/10/10 partition used in the COMD-YOLO manuscript submitted to Applied Intelligence. Contents: - splits/train_manifest.txt (2,400 stems) - splits/val_manifest.txt (300 stems) - splits/test_manifest.txt (300 stems) - splits/split_summary.json - annotation_audit.json - dataset_stats.json / dataset_stats.csv - cooccurrence_priors.json