DxPostprocess

DxPostprocess is a pipeline element that handles output tensors generated by the DxInfer element. It performs model-specific post-processing, such as classification, detection, or segmentation, and produces structured results by creating or modifying DXObjectMeta.

The component supports both built-in and user-defined logic, depending on the configured model type. Processed results are attached to DXFrameMeta, enabling downstream tasks such as visualization, event handling, or application-specific decisions.

Key Features

Modes of Operation

• Primary Mode: Creates new DXObjectMeta objects and attaches them to the corresponding DXFrameMeta
• Secondary Mode: Modifies existing DXObjectMeta based on updated inference results from DxInfer

Custom Postprocessing
You can use a custom post-processing function by specifying:

• library-file-path: Path to the shared object (.so) containing your custom function
• function-name: Name of the function to be called

The custom function signature has been updated to include GstBuffer access:

extern "C" void PostProcess(GstBuffer *buf,
                           std::vector<dxs::DXTensor> network_output,
                           DXFrameMeta *frame_meta,
                           DXObjectMeta *object_meta);

This allows full flexibility to implement custom decoding, filtering, or data transformation logic with direct access to the GStreamer buffer.

Hierarchy

GObject
 +----GInitiallyUnowned
       +----GstObject
             +----GstElement
                   +----GstBaseTransform
                         +----GstDxPostprocess

Properties

Name	Description	Type	Default Value
name	Sets the unique name of the DxPostprocess element.	String	"dxpostprocess0"
config-file-path	Path to the JSON config file containing the element's properties.	String	null
inference-id	Specifies the ID of the DxInfer output tensor to use for postprocessing.	Integer	0
secondary-mode	Enables Secondary Mode for processing object regions.	Boolean	false
library-file-path	Path to the custom preprocess library, if used.	String	null
function-name	Name of the custom preprocessing function to use.	String	null

Example JSON Configuration

{
    "inference_id": 1,
    "library_file_path": "/usr/local/share/gstdxstream/lib/libpostprocess_yolo.so",
    "function_name": "YOLOV5Pose_1"
}

NOTE

• All properties can also be configured using a JSON file for enhanced usability and flexibility.

• For implementing custom preprocess logic, refer to Chapter. Writing Your Own Application “Custom Post-Process Library Documentation”.

---