Skip to main content

DENK SDK API Reference

Overview​

The DENK SDK provides a set of tools for running neural network inference locally and offline. It supports both Python and C# and is designed to work with models created on the DENK Vision AI Hub.

Classes​

InferencePipeline​

The InferencePipeline class is the main entry point for setting up and running inference with the DENK SDK.

Constructor​

InferencePipeline(device: Device = Device.CPU, token: Union[str, None] = None)
  • Parameters:
    • device: The device to be used for inference. Defaults to Device.CPU.
    • token: The authentication token. If not provided, the function will look for a USB-dongle.

Methods​

  • get_backend_library() -> str

    • Returns the type and version of the backend library being used.

  • add_model(model_path: str) -> Model

    • Loads a model from the specified path and adds it to the pipeline.

  • get_models() -> List[Model]

    • Returns a list of models currently loaded in the pipeline.

  • run(image: np.ndarray) -> InferencePipeline

    • Runs inference on the provided image.

  • get_results()

    • Retrieves the results of the last inference run.

  • get_result_image(object_opacity: float = 0.5, pil_format: bool = False)

    • Returns the result image with specified object opacity and format.

Model​

Represents a model loaded into the InferencePipeline.

Constructor​

Model(parent_pipeline: InferencePipeline, model_path: str)
  • Parameters:
    • parent_pipeline: The InferencePipeline instance to which this model belongs.
    • model_path: The path to the model file.

Attributes​

  • pre: An instance of the Pre class for pre-processing settings.
  • post: An instance of the Post class for post-processing settings.

Pre​

Handles pre-processing settings for a model.

Methods​

  • set_evaluation_size(width: int, height: int) -> Pre

    • Sets the evaluation size for the model.

  • set_image_partitioning(partitions_in_width: int, partitions_in_height: int) -> Pre

    • Sets the image partitioning for the model.

  • set_moving_window(active: bool) -> Pre

    • Enables or disables the moving window feature.

Post​

Handles post-processing settings for a model.

Methods​

  • filter_by_confidence(minimum_confidence: float) -> Post

    • Filters results by a minimum confidence threshold.

  • set_segmentation_threshold(segmentation_threshold: float) -> Post

    • Sets the segmentation threshold for the model.

  • filter_overlapping_bounding_boxes(overlap_threshold: float) -> Post

    • Filters overlapping bounding boxes based on the specified threshold.

Enums​

Device​

An enumeration of available devices for running inference.

  • CPU = -1
  • GPU1 = 0
  • GPU2 = 1
  • GPU3 = 2
  • GPU4 = 3
  • GPU5 = 4
  • GPU6 = 5
  • GPU7 = 6
  • GPU8 = 7

Example Usage​

import denk_sdk

pipeline = denk_sdk.InferencePipeline(token="your_token", device=denk_sdk.Device.CPU)
model = pipeline.add_model("path/to/model.denk")

results = pipeline.run(image).get_results()
print(results)

Return Overview​

The returned protobuf object contains the following fields:

  • classification_models: List of classification models and their results.

    • model_id: Identifier of the model.
    • model_name: Name of the model.
    • run_time_ms: Time taken to run the model in milliseconds.
    • classes: List of classification results.
      • class_id: Identifier of the class.
      • class_name: Name of the class.
      • class_color: Color associated with the class.
      • classifier: Confidence score of the classification.

  • segmentation_models: List of segmentation models and their results.

    • model_id: Identifier of the model.
    • model_name: Name of the model.
    • run_time_ms: Time taken to run the model in milliseconds.
    • classes: List of segmentation results.
      • class_id: Identifier of the class.
      • class_name: Name of the class.
      • class_color: Color associated with the class.
      • objects: List of segmented objects.
        • bounding_box: Bounding box of the object.
        • oriented_bounding_box: Oriented bounding box of the object.
        • confidence: Confidence score of the segmentation.
        • segmentation: Segmentation mask.

  • instance_segmentation_models: List of instance segmentation models and their results.

    • Same structure as segmentation_models.

  • object_detection_models: List of object detection models and their results.

    • model_id: Identifier of the model.
    • model_name: Name of the model.
    • run_time_ms: Time taken to run the model in milliseconds.
    • classes: List of object detection results.
      • class_id: Identifier of the class.
      • class_name: Name of the class.
      • class_color: Color associated with the class.
      • objects: List of detected objects.
        • bounding_box: Bounding box of the object.
        • oriented_bounding_box: Oriented bounding box of the object.
        • confidence: Confidence score of the detection.

  • anomaly_detection_models: List of anomaly detection models and their results.

    • model_id: Identifier of the model.
    • model_name: Name of the model.
    • run_time_ms: Time taken to run the model in milliseconds.
    • classes: List of anomaly detection results.
      • class_id: Identifier of the class.
      • class_name: Name of the class.
      • class_color: Color associated with the class.
      • anomaly_score: Anomaly score.
      • objects: List of detected anomalies.
        • bounding_box: Bounding box of the anomaly.
        • oriented_bounding_box: Oriented bounding box of the anomaly.
        • confidence: Confidence score of the anomaly.
        • segmentation: Segmentation mask.

  • optical_character_recognition_models: List of OCR models and their results.

    • model_id: Identifier of the model.
    • model_name: Name of the model.
    • run_time_ms: Time taken to run the model in milliseconds.
    • text: Recognized text.
    • confidence: Confidence score of the OCR.

  • barcode_reading_models: List of barcode reading models and their results.

    • model_id: Identifier of the model.
    • model_name: Name of the model.
    • run_time_ms: Time taken to run the model in milliseconds.
    • classes: List of barcode reading results.
      • class_id: Identifier of the class.
      • class_name: Name of the class.
      • class_color: Color associated with the class.
      • barcode_type: Type of the barcode.
      • objects: List of detected barcodes.
        • bounding_box: Bounding box of the barcode.
        • oriented_bounding_box: Oriented bounding box of the barcode.
        • text: Decoded text from the barcode.
        • bytes: Raw bytes of the barcode.

Example Return​

{
    "classification_models": [
        {
            "model_id": "model_1",
            "model_name": "Classification Model 1",
            "run_time_ms": 123.45,
            "classes": [
                {
                    "class_id": "class_1",
                    "class_name": "Class 1",
                    "class_color": {"red": 255, "green": 0, "blue": 0},
                    "classifier": 0.95
                }
            ]
        }
    ],
    "segmentation_models": [
        {
            "model_id": "model_2",
            "model_name": "Segmentation Model 1",
            "run_time_ms": 234.56,
            "classes": [
                {
                    "class_id": "class_2",
                    "class_name": "Class 2",
                    "class_color": {"red": 0, "green": 255, "blue": 0},
                    "objects": [
                        {
                            "bounding_box": {"top_left": {"x": 10, "y": 20}, "bottom_right": {"x": 30, "y": 40}},
                            "oriented_bounding_box": {"by_center": {"center": {"x": 20, "y": 30}, "width": 20, "height": 20, "angle": 0, "full_orientation": True}},
                            "confidence": 0.85,
                            "segmentation": b'\x89PNG\r\n\x1a\n...'
                        }
                    ]
                }
            ]
        }
    ]
}