Features - Features - 6.3 English - XD056

Supports various VVAS Gstreamer components.

• vvas_xinfer : performs inference on input video images on hardware accelerators or CPU for real-time analytics.

• vvas_xabrscaler : performs scaling and format conversion. Also used with vvas_xinfer for external pre-processing.

• vvas_xmetaconvert : performs transforming of VVAS ML inference metadata into overlay metadata for the vvas_xoverlay

• vvas_xoverlay : overlays the inference results on the input video stream for real-time visualization.

• vvas_xmulticrop : performs resizing, color space conversion, preprocessing, and cropping of multiple objects from a single input frame for machine learning applications.

• vvas_xmetaaffixer : adjusts inference metadata for different resolutions, ensuring compatibility with original frames by using master and slave pads to align metadata across multiple input sources.

• vvas_xfilter : an infrastructure plugin to enable any hardware/software kernel to be used into VVAS Gstreamer pipeline

Supports inference using ONNX Runtime with Vitis-AI Execution Provider, ONNX RT with CPU Execution Provider and VART.

Facilitates execution of models that are partially offloaded, ensuring completion using the ONNX Runtime’s default CPU Execution Provider.

VART backend can be used for models that are fully offloaded to AI engine.

The vvas_xinfer supports VART inference using zero-copy mode for optimal performance.

VVAS offers support for popular AI model classes like Object Detection, Object Classification, Segmentation, Pose Detection etc. (This release is tested with resnet classification models.)

Supports input from file source.

Provides hardware-accelerated resizing and color space conversion.

Supports overlaying bounding boxes, text, circles, arrows, polygons, and more around detected objects.

vvas_xinfer includes built-in hardware acceleration for pre-processing to convert incoming data to compatible formats and layouts.

Offers post-processing support for ResNet50 and ResNet18 models.

vvas_xinfer supports standard video formats and VVAS-specific tensor formats. For the custom tensor format names shown in caps and debug logs, see Custom AI Tensor Formats.