ONNXModelHub

class ONNXModelHub(cacheDirectory: File) : ModelHub

This class provides methods for loading ONNX model to the local cacheDirectory.

Since

0.3

Parameters

cacheDirectory

The directory for all loaded models. It should be created before model loading and should have all required permissions for file writing/reading on your OS.

Constructors

ONNXModelHub

jvm

fun ONNXModelHub(cacheDirectory: File)

Functions

get

jvm

operator fun <T : InferenceModel, U : InferenceModel> get(modelType: ModelType<T, U>): U

loadModel

jvm

open override fun <T : InferenceModel, U : InferenceModel> loadModel(modelType: ModelType<T, U>, loadingMode: LoadingMode): T

Loads model configuration without weights.

fun loadModel(modelType: OnnxModelType<*>, vararg executionProviders: ExecutionProvider, loadingMode: LoadingMode = LoadingMode.SKIP_LOADING_IF_EXISTS): OnnxInferenceModel

This method loads model from ONNX model zoo corresponding to the specified modelType. The loadingMode parameter defines the strategy of existing model use-case handling. If loadingMode is LoadingMode.SKIP_LOADING_IF_EXISTS and the model is already loaded, then the model will be loaded from the local cacheDirectory. If loadingMode is LoadingMode.OVERRIDE_IF_EXISTS the model will be overridden even if it is already loaded. executionProviders is a list of execution providers which will be used for model inference.

loadPretrainedModel

jvm

fun <T : InferenceModel, U : InferenceModel> loadPretrainedModel(modelType: ModelType<T, U>, loadingMode: LoadingMode): U

Properties

cacheDirectory

jvm

val cacheDirectory: File