Grounding DINO Prelabeling

src.pipeline.prelabelling.grounding_dino_prelabelling.generate_gd_prelabelling(raw_dir: Path, output_dir: Path, config: Dict, model_weights: Path, config_path: Path, text_prompts: List[str] = ['fire', 'smoke', 'person', 'vehicle', 'lightning'], box_threshold: float = None, text_threshold: float = None) → None

Run Grounding DINO to detect objects from images using text prompts. Saves a JSON file per image with predicted bounding boxes and metadata. This version runs sequentially (no multiprocessing).