image_select/full/train_eval.json

{
  "_note": "full mode: each of the 9 tiles is an independent photo. N positives + (9-N) negatives per reset; both dirs are relative to assets/image_select/.",
  "content": {
    "positive_dir": "full/crosswalk/positive",
    "negative_dir": "full/crosswalk/negative",
    "target_label": "crosswalks",
    "pos_count_range": [2, 4]
  },
  "render": {
    "tile_display_px": 150,
    "grid_gap_px": 4
  }
}