Human-in-the-Loop Module

Script for human review of the unmatched YOLO and GroundingDINO results.

src.pipeline.human_intervention.export_versioned_results(project_id: str, output_dir: Path, version: str = None) → dict[source]

Export human review results to a versioned file.

Parameters:

  • project_id – Label Studio project identifier.

  • output_dir – Directory to save human review results.

  • version – Version string for the output file.

Returns:

Human review results.

src.pipeline.human_intervention.import_tasks_to_project(base_url: str, headers: Dict[str, str], project_id: int, tasks_file: Path) → bool[source]

Import tasks from a JSON file into a Label Studio project.

Loads pre-annotated tasks into the given project for human review.

Parameters:

  • base_url – Label Studio server URL.

  • headers – HTTP headers with the API token.

  • project_id – ID of the target project.

  • tasks_file – Path to the JSON file with tasks.

Returns:

True if import succeeded, False otherwise.

Return type:

bool

src.pipeline.human_intervention.run_human_review(project_name: str = 'AutoML-Human-Intervention', export_results_flag: bool = None) → dict[source]

Run the complete human review workflow.

This function processes JSON files from the pending directory, creates a Label Studio project, and handles the human review workflow.

Parameters:

  • project_name – Name of the Label Studio project

  • export_results_flag – Whether to export results (None = ask user)

Returns:

Dictionary containing the results of the human review process.

Return type:

dict

src.pipeline.human_intervention.setup_label_studio(project_name: str, output_dir: str) → dict[source]

Setup a Label Studio project for human review.

Parameters:

  • project_name – Name of the Label Studio project.

  • output_dir – Directory to store Label Studio data.

Returns:

Dictionary with project configuration details.

src.pipeline.human_intervention.transform_reviewed_results_to_labeled(export_results: list, labeled_dir: Path = None, image_dir: Path = None) → int[source]

Transform reviewed Label Studio results to labeled directory. Removes original pending files and saves human-reviewed data.

Parameters:

  • export_results (list) – List of results from Label Studio export

  • labeled_dir (Path) – Directory to save labeled files

  • image_dir (Path) – Directory containing images

Returns:

Number of files transformed

Return type:

int