ml Package

data_utils Module

ml.data_utils.coco_image_data(coco, img_id)

Extract image data and annotations from COCO object.

Args:

coco: COCO object. img_id: Image ID.

Returns:

Tuple[str, dict]: Tuple containing image path and annotation dictionary.

ml.data_utils.image_predictions_to_id(image_path, model_version, image_shape, prediction, classes, dataset_id)

Convert image predictions to a specific format.

Args:

image_path: Path to the image. model_version: Model version. image_shape: Shape of the image. prediction: Model prediction. classes: List of classes. dataset_id: id of label studio project with data

Returns:

dict: Serialized prediction.

ml.data_utils.load_coco_from_json(path)

Load COCO object from a JSON file.

Args:

path (str): Path to the JSON file.

Returns:

pycocotools.coco.COCO: COCO object.

ml.data_utils.load_coco_from_prediction(prediction)

Load COCO object from a prediction.

Args:

prediction: Prediction data. categories: List of categories. image_data: Image data.

Returns:

pycocotools.coco.COCO: COCO object.

ml.data_utils.parse_image_json(label_data)

Parse JSON data for image information.

Args:

label_data: JSON data containing image information.

Returns:

Tuple[list, list]: Tuple containing lists of image paths and annotations.

ml.data_utils.parse_predictions(image_shape, prediction, classes_by_id)

Parse model predictions.

Args:

image_shape: Shape of the image. prediction: Model prediction. classes_by_id: Mapping of class IDs to class names.

Returns:

List[dict]: List of parsed predictions.

ml.data_utils.sample_prediction_to_dict(image_shape, bbox, label, classes_by_id)

Convert a sample prediction to a dictionary.

Args:

image_shape: Shape of the image. bbox: Bounding box coordinates. label: Label of the prediction. classes_by_id: Mapping of class IDs to class names.

Returns:

dict: Serialized sample prediction.

ml.data_utils.serialize_predictions(image_shapes, predictions, image_paths, model_version, classes, dataset_id)

Serialize predictions into a specific format.

Args:

image_shapes: Shapes of the images. predictions: Model predictions. image_paths: Paths to the images. model_version: Model version. classes: List of classes. dataset_id: id of label studio project with data

Returns:

List: List of serialized predictions.

detector_dataset Module

class ml.detector_dataset.DetectorDataset(dataset_path, annotation_path, transforms=None, indices=None)

Dataset class for object detection.

Args:

dataset_path (str): Path to the dataset. annotation_path (str): Path to the annotation file. transforms (callable, optional): Optional transforms to be applied to the data. indices (list, optional): List of indices to subset the dataset.

Attributes:

indices (list): List of indices to subset the dataset. dataset_path (str): Path to the dataset. annotation_path (str): Path to the annotation file. transforms (callable, optional): Optional transforms to be applied to the data. img_paths (list): List of image paths. targets (list): List of targets. coco (pycocotools.coco.COCO): COCO object for indexing.

Methods:

__getitem__: Get a specific item from the dataset. __len__: Get the length of the dataset.

detector_module Module

class ml.detector_module.CocoDetectorABC(num_classes, iou_thresh, **kwargs)

Abstract base class for the COCO Detector.

Args:

num_classes (int): Number of classes in the dataset. iou_thresh (float): IoU threshold for non-maximum suppression. **kwargs: Additional keyword arguments.

Attributes:

num_classes (int): Number of classes. iou_thresh (float): IoU threshold for NMS. model: Model to be initialized in subclasses.

forward(x, y)

Forward pass of the model.

Args:

x: Input tensor. y: Ground truth tensor.

Returns:

Model output.

predict(x)

Make predictions using the model.

Args:

x: Input tensor.

Returns:

list: List of final predictions after NMS.

class ml.detector_module.CocoDetectorModule(model: CocoDetectorABC)

Lightning Module for training, validation, and testing of the COCO Detector.

Args:

model (CocoDetectorABC): COCO Detector model.

Attributes:

model (CocoDetectorABC): COCO Detector model.

configure_optimizers()

Configure the optimizer for the Lightning Module.

Returns:

torch.optim.Optimizer: The optimizer for the model training.

predict(x) → Any

Perform predictions using the model.

Args:

x: Input data.

Returns:

List[dict]: List of final predictions after NMS.

test_step(batch, batch_idx) → Tensor | Mapping[str, Any] | None

Test step for Lightning Module.

Args:

batch: Batch of test data. batch_idx (int): Batch index.

Returns:

torch.Tensor: Loss value for testing.

training_step(batch, batch_idx)

Training step for Lightning Module.

Args:

batch: Batch of training data. batch_idx (int): Batch index.

Returns:

torch.Tensor: Loss value for training.

validation_step(batch, batch_idx) → Tensor | Mapping[str, Any] | None

Validation step for Lightning Module.

Args:

batch: Batch of validation data. batch_idx (int): Batch index.

Returns:

torch.Tensor: Loss value for validation.

class ml.detector_module.CocoDetectorResnet(num_classes, iou_thresh, **kwargs)

Subclass of CocoDetectorABC for ResNet-based object detection model.

Args:

num_classes (int): Number of classes in the dataset. iou_thresh (float): IoU threshold for non-maximum suppression. **kwargs: Additional keyword arguments.

model_endpoint Module

ml.model_endpoint.annotation_path_from_id(model_id: int)

Get the annotation path based on the model ID.

Args:

model_id (int): ID of the model.

Returns:

str: Annotation path.

ml.model_endpoint.dataset_path_from_id(project_id: int)

Get the dataset path based on the project ID.

Args:

project_id (int): ID of the project.

Returns:

str: Dataset path.

ml.model_endpoint.evaluate_detector(model_id: int, model: CocoDetectorModule, dataset: DetectorDataset) → list[float]

Evaluate the object detection model using COCO evaluation.

Args:

model_id (int): ID of the model. model (CocoDetectorModule): Object detection model. dataset (DetectorDataset): COCO dataset for evaluation.

Returns:

list: COCO evaluation statistics.

ml.model_endpoint.image_paths_for_prediction(project_id: int, model_id: int)

Get unlabeled image paths for prediction.

Args:

project_id (int): ID of the project. model_id (int): ID of the model.

Returns:

set[str]: Unlabeled image paths.

ml.model_endpoint.initialize_coco_loader(dataset_path: str, annotation_path: str, batch_size: int, is_shuffled: bool = False, indices: list | None = None)

Initialize COCO DataLoader for the object detection dataset.

Args:

dataset_path (str): Path to the dataset. annotation_path (str): Path to the annotation file. batch_size (int): Batch size. is_shuffled (bool): Flag for shuffling the dataset. indices (list): List of indices to subset the dataset.

Returns:

DataLoader: COCO DataLoader.

ml.model_endpoint.initialize_model(model_id: int)

Initialize the object detection model.

Args:

model_id (int): ID of the model.

Returns:

CocoDetectorResnet: Initialized object detection model.

ml.model_endpoint.initialize_optimizer(model: CocoDetectorABC, model_id: int)

Initialize the optimizer for the object detection model.

Args:

model: Object detection model. model_id (int): ID of the model.

Returns:

torch.optim.AdamW: Initialized optimizer.

ml.model_endpoint.load_label_studio_project(project_id: int)

Load label studio project id from the database.

Args:

project_id (int): ID of the project.

Returns:

Int: Label studio project id.

ml.model_endpoint.load_model_params(model_id: int)

Load model parameters from the database.

Args:

model_id (int): ID of the model.

Returns:

LearningModel: Model parameters.

ml.model_endpoint.predict(project_id: int, model_id: int) → dict

Train the object detection model.

Args:

project_id (int): ID of the project. model_id (int): ID of the model.

Returns:

dict: COCO prediction dictionary

ml.model_endpoint.project_classes(model_id: int)

Get the list of classes from the annotation file of the model.

Args:

model_id (int): ID of the model.

Returns:

list: List of class names.

ml.model_endpoint.train(project_id: int, model_id: int) → list

Train the object detection model.

Args:

project_id (int): ID of the project. model_id (int): ID of the model.

Returns:

list: COCO evaluation statistics.