pixaris.orchestration package

Submodules

pixaris.orchestration.base module

pixaris.orchestration.base.generate_image(data, image_generator, args, failed_args)[source]

Generates a single image based on the provided data and image generator.

Parameters:

  • data (dict) – input data, e.g. a dictionary containing images and masks

  • image_generator (ImageGenerator) – the image generator to use for generating images

  • args (dict) – additional arguments for image generation. To be set during generation

  • failed_args (list) – list to store failed arguments. Has to exist and be handled outside this function.

Returns:

generated image and name

Return type:

tuple[PIL.Image.Image, str]

pixaris.orchestration.base.generate_images_based_on_dataset(data_loader: DatasetLoader, image_generator: ImageGenerator, experiment_handler: ExperimentHandler, metrics: list[BaseMetric], args: dict[str, any]) → Iterable[tuple[Image, str]][source]

Generates images based on an evaluation set. This function loads a dataset using the provided data loader, generates images using the provided image generator, and stores the results using the provided experiment handler.

Parameters:

  • data_loader (DatasetLoader) – An instance of DatasetLoader to load the dataset.

  • image_generator (ImageGenerator) – An instance of ImageGenerator to generate images.

  • experiment_handler (ExperimentHandler) – An instance of ExperimentHandler to store the generated images and results.

  • metrics (list[BaseMetric]) – A list of metrics to calculate.

  • args (dict[str, any]) – A dictionary of arguments to be used for image generation and storing results.

Returns:

A list of generated images and names

Return type:

list[tuple[PIL.Image.Image, str]]

pixaris.orchestration.base.generate_images_for_hyperparameter_search_based_on_dataset(data_loader: DatasetLoader, image_generator: ImageGenerator, experiment_handler: ExperimentHandler, metrics: list[BaseMetric], args: dict[str, any])[source]

Generates images for hyperparameter search based on the evaluation set. This function performs a hyperparameter search by generating images for each combination of hyperparameters provided. It validates the hyperparameters, generates a grid of hyperparameter combinations, and then generates images for each combination using the provided data loader, image generator, and experiment handler.

Parameters:

  • data_loader (DatasetLoader) – The data loader to load the evaluation set.

  • image_generator (ImageGenerator) – The image generator to generate images.

  • experiment_handler (ExperimentHandler) – The experiment handler to save generated images.

  • metrics (list[BaseMetric]) – A list of metrics to calculate.

  • args – A dictionary of arguments, including:

  • “workflow_apiformat_json” (str): The path to the workflow file in API format.

  • “hyperparameters” list(dict): A dictionary of hyperparameters to search. Each element of the list should be compatible with the generation parameters, that the image_generator takes as an input.

  • “experiment_run_name” (str): The base name for each run.

Raises:

ValueError – If no hyperparameters are provided or if the hyperparameters are invalid.

pixaris.orchestration.kubernetes module

pixaris.orchestration.kubernetes.copy_bytes_to_pod(kube_conn, namespace: str, pod_name: str, bytes_to_copy: bytes, dest_path: str)[source]

Copy a file to a pod comparable with kubectl cp. See here: https://github.com/kubernetes-client/python/issues/476#issuecomment-2701387338

Parameters:

  • kube_conn – The connection to the Kubernetes cluster.

  • namespace – The namespace of the pod.

  • pod_name – The name of the pod.

  • bytes_to_copy – The bytes to be copied.

  • dest_path – The destination path on the pod.

pixaris.orchestration.kubernetes.pixaris_orchestration_kubernetes_locally(data_loader: DatasetLoader, image_generator: ImageGenerator, experiment_handler: ExperimentHandler, metrics: list[BaseMetric], args: dict[str, any], auto_scale: bool)[source]

Trigger remote evaluation on the Kubernetes cluster. This function will pickle the inputs and upload them to the Kubernetes cluster. It will then trigger the remote evaluation.

Parameters:

  • data_loader (DatasetLoader) – The data loader to load the evaluation set.

  • image_generator (ImageGenerator) – The image generator to generate images. E.g. ComfyClusterGenerator

  • experiment_handler (ExperimentHandler) – The experiment handler to save generated images.

  • metrics (list[BaseMetric]) – The metrics to calculate.

  • auto_scale (bool) – Whether to auto scale the cluster to the maximum number of parallel jobs.

  • args – A dictionary of arguments, including:

  • “workflow_apiformat_json” (str): The path to the workflow file in API format.

  • “workflow_pillow_image” (PIL.Image): The image to use as input for the workflow.

  • “dataset” (str): The evaluation set to use.

  • “pillow_images” (list[dict]): A list of images to use as input for the workflow.

  • “experiment_run_name” (str): The name of the run.

  • “max_parallel_jobs” (int): The maximum number of parallel jobs to run.

Module contents