splits/caltech256/split_config.json

{
  "chosen_raw_source": "data/caltech256",
  "class_decision": "Use 256 object classes by default; clutter is excluded explicitly and recorded in split_config.json.",
  "dataset": "caltech256",
  "dry_run": false,
  "duplicate_policy": "Same-label groups retain one deterministic representative; every member of a cross-label group is excluded before splitting.",
  "excluded_classes": [
    "257.clutter"
  ],
  "excluded_cross_label_conflict_files": 4,
  "excluded_same_label_duplicate_files": 13,
  "include_clutter": false,
  "interpolation": "bilinear",
  "link_mode": "copy",
  "ratios": {
    "test": 0.15,
    "train": 0.7,
    "val": 0.15
  },
  "raw_object_image_count": 29780,
  "records": 29763,
  "split_policy": "exclude_clutter_and_exact_duplicate_conflicts_then_stratified_70_15_15",
  "split_seed": 42,
  "target_size": 224,
  "usable_image_count": 29763,
  "warnings": []
}