DxOsd

DxOsd is a GStreamer element that provides On-Screen Display (OSD) capabilities by overlaying object information onto the original video frame.
It uses object metadata (Object Meta) passed from upstream elements, such as DxPostprocess or DxTracker, to render visual elements directly on the frame.

Key Features

Draw Inference Results

• Draws bounding boxes on video frames
• Displays class labels and confidence scores
• Colors boxes by track ID or class
• Supports segmentation maps, human pose, and facial landmarks

Hierarchy

GObject
 +----GInitiallyUnowned
       +----GstObject
             +----GstElement
                   +----GstBaseTransform
                         +----GstDxOsd

Pad Templates

Sink (input)

Property	Value
Format	video/x-raw, format=(string){ RGB, BGR, NV12, I420 }; application/x-dxvideoraw

Src (output)

Property	Value
Format	video/x-raw, format=(string){ RGB, BGR, NV12, I420 }; application/x-dxvideoraw

Domain Mode Behavior

dxosd is a dual-mode element (see Multi-Stream Domain).

• Normal mode (sink caps = video/x-raw): a single video_info is taken from the caps and reused for every buffer.
• Domain mode (sink caps = application/x-dxvideoraw): dxosd reads the per-stream wrapped CAPS events emitted by dxinputselector (one per stream, carrying the original video/x-raw caps and the stream-id) and stores a separate video_info per _stream_id. For each incoming buffer it picks the matching video_info from DXFrameMeta._stream_id, so streams with different resolutions or formats are rendered correctly.

Properties

Name	Description	Type	Default Value
name	Sets the unique name of the DxOsd element.	String	"dxosd0"

NOTE

• DXOSD supports RGB, BGR, NV12, and I420 input formats. The output format matches the input format (passthrough — no format conversion is applied).
• In DXOSD's output buffers, the DXFrameMeta metadata is preserved and passed to downstream elements.
• Visualizations include bounding boxes, class names, confidence scores, and additional data like segmentation maps, poses, or face landmarks, depending on available metadata.

Multi-Stream Support

DxOsd can be safely placed downstream of DxInputSelector. It automatically tracks per-stream video information via wrapped caps events, allowing correct OSD rendering for buffers from different streams with different resolutions and formats.

---