segpath/sampling_config.json

{
  "package": "segpath",
  "data_role": "optional_auxiliary_training",
  "enabled_by_default": false,
  "default_mix_probability": 0.0,
  "merged_with_main_package": false,
  "sampling_order_if_enabled": ["marker", "parent_query", "retained_target"],
  "reason": "small semantic-region targets; dataset size must not dominate main ROI training"
}