splits/cifar100/split_config.json

{
  "compatibility_copy": "data/processed/cifar100/cifar-100-python",
  "cross_label_identical_image_groups": 17,
  "dataset": "cifar100",
  "dry_run": false,
  "duplicate_anomalies_used_to_modify_benchmark": false,
  "duplicate_policy": "Diagnostic only. Official CIFAR samples, labels, and official train/test membership are never changed because of exact-image anomalies.",
  "link_mode": "copy",
  "official_test_count": 10000,
  "official_test_preserved_unchanged": true,
  "official_train_count": 50000,
  "official_train_test_duplicate_groups": 10,
  "records": 60000,
  "same_label_duplicate_groups": 9,
  "split_policy": "official_train_stratified_450_50_per_class_official_test_untouched",
  "split_seed": 42,
  "test_count": 10000,
  "test_policy": "Official CIFAR-100 test pickle is untouched and is never used for validation/checkpoint selection.",
  "torchvision_root": "data/processed/cifar100",
  "train_count": 45000,
  "val_count": 5000,
  "warnings": []
}