DxMsgConv

DxMsgConv is an element that processes inference metadata from upstream DxPostprocess elements and converts it into message payloads in various formats.

This element bridges internal AI pipeline processing and external communication systems. It receives the raw inference metadata—data about detected objects, classifications, and tracking—from the upstream DxPostprocess elements. DxMsgConv then transforms this complex metadata into structured message payloads using various standard formats, such as JSON or Protocol Buffers. Once converted, these payloads are passed downstream to the DxMsgBroker element, which is responsible for transmitting them reliably to an external broker server (like an MQTT or Kafka broker).

To properly implement the specific structure and content required for the final messages—referred to as the "actual message format logic"—a user-defined custom library is required. This library contains the unique code that dictates how the inference metadata is mapped into the desired message payload structure. For guidance on developing and integrating this custom logic, users must consult the relevant documentation, specifically: Chapter. Writing Your Own Application "Custom Message Convert Library Documentation".

Key Features

Metadata Conversion

• Converts metadata generated by DxPostprocess into structured message formats through custom library implementation.
• The actual message format and structure are defined by the user-implemented custom library.
• Requires a custom shared library to define and implement the required message formats.

Integration with DxMsgBroker

• Passes formatted message payloads (as generated by the custom library) downstream to DxMsgBroker, which handles delivery to external systems.

Configuration

• Message format conversion can be customized using a configuration file.

Example Output Format

The element processes metadata through the custom library to generate message payloads. The default example custom library provided with DX-STREAM generates JSON messages with comprehensive object detection information:

{
  "streamId": 0,
  "seqId": 123,
  "width": 1920,
  "height": 1080,
  "objects": [
    {
      "object": {
        "label_id": 1,
        "track_id": 42,
        "confidence": 0.87,
        "name": "person",
        "box": { ... },
        "body_feature": [...],
        "segment": { ... },
        "pose": { ... },
        "face": { ... }
      }
    }
  ]
}

NOTE

The actual message format depends entirely on the implementation of your custom library. The above is just an example from the provided default library. For detailed implementation guidance and complete JSON structure examples, refer to Chapter 4. Writing Your Own Application.

In the default library, frame-level segmentation remains a full-frame semantic class map, while object-level segment payloads are ROI-local binary masks aligned to the corresponding object.box.

Hierarchy

GObject
 +----GInitiallyUnowned
       +----GstObject
             +----GstElement
                   +----GstBaseTransform
                         +----GstDxMsgConv

Pad Templates

Sink (input)

Property	Value
Format	video/x-raw; application/x-dxvideoraw

Src (output)

Property	Value
Format	video/x-raw; application/x-dxvideoraw

DxMsgConv is essentially caps-agnostic — it does not inspect raw video pixels. It accepts either video/x-raw or application/x-dxvideoraw solely so it can be placed anywhere in the chain; processing operates on the buffer's DXFrameMeta / DXObjectMeta only.

Custom Library Error Reporting

Custom message-converter libraries must follow the error reporting contract described in Chapter. Writing Your Own Application "Error Reporting from Custom Libraries" — no g_error() / abort(); use g_warning() + return nullptr for per-frame skips, and throw std::runtime_error for permanent errors.

Properties

Name	Description	Type	Default Value
config-file-path	Path to the configuration file. (optional).	String	null
library-file-path	Path to the custom message converter library. Required.	String	null
message-interval	Frame interval at which message is converted.	Integer	1
include-frame	Flag whether to include frame data as base64 JPEG in the message. (optional).	Boolean	false

Limitation

DxMsgConv must not be placed after an input-selector element. Caps are configured once via set_caps callback and are not re-negotiated during stream switching. Placing DxMsgConv downstream of input-selector may result in caps mismatch when streams are switched at runtime.

---

NOTE

• The element itself does not define the message format - this is entirely determined by your custom library implementation.
• The default example library provided with DX-STREAM generates JSON format, but you can implement any message format in your custom library.

---