active_learning_queue_rank

Rank unlabeled images by uncertainty using a YOLO checkpoint and populate the review queue.

Inputs

• checkpoint: Path to the YOLO model checkpoint weights.
• unlabeled_pool: Path to the folder containing unlabeled images.
• dataset_path: Path to the destination YOLO dataset directory.
• strategy: Strategy for scoring uncertainty (entropy, margin, least_confident).
• limit: Maximum number of images to rank.
• imgsz: Inference image size.
• conf: Confidence threshold.
• device: Device to run inference on.

Outputs

• ranked_count: Number of images processed and added.
• queue_entries: List of review queue entries.

Examples

{
  "checkpoint": "yolov8n.pt",
  "unlabeled_pool": "pool",
  "dataset_path": "dataset"
}