NPU

The following nodes can only be used in combination with a VIU2.

Classify Image (NPU)

The "Classify Image" node (NPU) executes an image classification model directly on a neural processing unit (NPU). The NPU analyzes an input image and assigns it to a predefined class. Hardware acceleration on the NPU enables particularly fast and energy-efficient inference, making it ideal for mobile devices, edge systems, and embedded applications.

Flow

Parameter set

Attribute	Type	Description
Classification Component	NPU Yolo Classification	Requires the Get node of the NPU Yolo Classification component.
Images	Image <Array>	One or more images are expected.
Classes	String <Array>	All classes (sorted by Confidence) that are potential classes.
Messages	String <Array>	An array of messages that describe the result of evaluating each input image.
Raw Results	Inference Classification Result <Array>	Includes all classes (sorted by confidence) that are potential classes, the corresponding confidence values, the pre-processed and post-processed image, and whether and as what the image was classified.

Detect Objects (NPU)

The "Detect Objects" (NPU) node uses the NPU to recognize multiple objects in an image and determine their positions (boundary frames). In addition to class assignment, the model also provides the coordinates of the detected objects. For NPUs, it is important to use optimized models (e.g., small YOLO variants) and to correctly configure post-processing, such as non-maximum suppression (NMS).

Flow

Parameter set

Attribute	Type	Description
Detection Component	NPU Yolo Detection	Requires the Get node of the NPU Yolo Detection component.
Images	Image <Array>	One or more images are expected.
Object Names	String <Array>	Lists the names of the detected objects.
Messages	String <Array>	An array of messages that describe the result of evaluating each input image.
Raw Results	Inference Detection Result <Array>	In addition to an array of recognized objects, it also includes the pre- and post-processed image, information about how many objects were recognized, and whether any objects were found at all.

Segment Instances (NPU)

The "Segment Instances (NPU)" node extends object recognition by generating a pixel-accurate mask for each detected object. This identifies not only what is in the image and where, but also precisely which pixels belong to the object. Because this task is computationally intensive, NPUs often require the use of lighter models, and the outputs must then be further processed (e.g., smoothing or scaling of the masks).

Flow

Parameter set

Attribute	Type	Description
Segmentation Component	NPU Yolo Segmentation	Requires the Get node of the NPU Yolo Segmentation component.
Images	Image <Array>	One or more images are expected.
Object Names	String <Array>	Lists the names of the detected objects.
Messages	String <Array>	An array of messages that describe the result of evaluating each input image.
Raw Results	Inference Segmentation Result <Array>	Includes a list of segmentation objects as well as the pre- and post-processed image. Additionally, information is returned about how many objects were detected and whether any objects were found at all.