Run full preprocessing pipeline: cleaning + augmentation. Args: images_dir: Source images directory labels_d

Augment dataset by creating additional transformed versions of images and labels. Args: images_dir: Source images direct

Draw bounding boxes on image. Args: image: Input image as numpy array bboxes: List of bbox dicts class_names: Option

Load the YOLO model from the specified path.

Parse YOLO format labels and return bbox data in pixel coordinates. Args: label_path: Path to .txt label file img_width: Ima

Train the YOLO model.

Build Albumentations augmentation pipeline.

Load preprocessing configuration from YAML file.

Recursively merge update config into base config.

Clean dataset by removing corrupted files and invalid annotations. Args: images_dir: Directory containing images

Compute statistics for the dataset. Args: dataset_splits: Dictionary of split names to image paths labels_dir: Directory con

Create a default preprocessing configuration file. Args: output_path: Path to save the config file

Create the data.yaml file for YOLO training.

Annotate images using GroundingDINO and create a YOLO-compatible dataset. Args: input_folder (str): Path to folder containing input

Detect train/val/test images and labels paths in the dataset.

Download the Kaggle dataset.

Export the trained YOLO model to NCNN format.

Get sample images with annotations for display. Args: dataset_splits: Dataset splits sample_size: Number of images to sample

Perform inference on a single image.

Perform inference on a video file.

Perform real-time inference on webcam feed.

Load dataset structure and return image paths per split. Args: dataset_path: Path to dataset root directory Returns: Di

Initialize preprocessor with configuration. Args: config_path: Path to YAML config file. If None, uses defaults.