# Elements Links under the GStreamer element name (first column of the table) contain the description of element properties, in the format generated by gst-inspect-1.0 utility. ## Inference plugins | Element | Description | |----------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | [gvadetect](./gvadetect.md) | Performs object detection and *optionally* object classification/segmentation/pose estimation. Inputs: ROIs (regions of interest) or full frame. Output: object bounding box detection along with prediction metadata. The `queue` element must be put directly after the `gvadetect` element in pipeline.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect model=$mDetect device=GPU[CPU,NPU] ! queue ! … OUT
| | [gvaclassify](./gvaclassify.md) | Performs object classification/segmentation/pose estimation. Inputs: ROIs or full frame. Output: prediction metadata. The `queue` element must be put directly after the `gvaclassify` element in the pipeline.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect model=$mDetect device=GPU ! queue ! gvaclassify model=$mClassify device=CPU ! queue ! … OUT
| | [gvainference](./gvainference.md) | Executes any inference model and outputs raw results. Does not interpret data and does not generate metadata. The `queue` element must be put directly after the `gvainference` element in the pipeline.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect model=$mDetect device=GPU ! queue ! gvainference model=$mHeadPoseEst device=CPU ! queue ! … OUT
| | [gvatrack](./gvatrack.md) | Tracks objects across video frames using zero-term or short-term tracking algorithms. Zero-term tracking assigns unique object IDs and requires object detection to run on every frame. Short-term tracking allows for tracking objects between frames, reducing the need to run object detection on each frame.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect model=$mDetect device=GPU ! gvatrack tracking-type=short-term-imageless ! … OUT
| | [gvaaudiodetect](./gvaaudiodetect.md) | Legacy plugin. Performs audio event detection using the `AclNet` model.
Example:
gst-launch-1.0 … ! decodebin3 ! audioresample ! audioconvert ! audio/x-raw … ! audiomixer … ! gvaaudiodetect model=$mAudioDetect ! … OUT
| [gvaaudiotranscribe](./gvaaudiotranscribe.md) | ASR plugin. Performs audio transcription using `Whisper` model.
Example:
gst-launch-1.0 … ! decodebin3 ! audioresample ! audioconvert ! audio/x-raw … ! audiomixer … ! gvaaudiotranscribe model=$mASR device=CPU ! … OUT
| | [gvagenai](./gvagenai.md) | Performs inference using GenAI models. It can be used to generate text descriptions from images or video.
Example:
gst-launch-1.0 … ! decodebin3 ! videoconvert ! gvagenai model=$mGenAI device=GPU ! … OUT
| ## 3D plugins | Element | Description | |----------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | [g3dradarprocess](./g3dradarprocess.md) | Processes millimeter-wave (mmWave) radar signal data. Performs data reordering, pre-processing, DC (Direct Current) removal, and interfaces with the radar library to generate point clouds, clusters, and tracking data. Attaches custom metadata containing detected reflection points, clustered objects, and tracked targets to each buffer.
Example:
gst-launch-1.0 multifilesrc location=radar/%06d.bin ! application/octet-stream ! g3dradarprocess radar-config=config.json frame-rate=10 ! fakesink
| | [g3dlidarsrc](./g3dlidarsrc.md) | Captures real-time point clouds from a physical LiDAR device and emits `application/x-lidar` buffers with attached `LidarMeta`. A vendor-agnostic live source: a JSON config selects the vendor SDK backend (RoboSense today, loaded at runtime via `dlopen`), the device model, and the UDP transport. Its output is byte-for-byte compatible with `g3dlidarparse`, so it can drive `g3dinference` and the rest of a 3D pipeline unchanged.
Example:
gst-launch-1.0 g3dlidarsrc config=configs/robosense_e1r_udp.json ! g3dinference config=pointpillars_ov_config.json device=CPU ! gvametaconvert format=json ! gvametapublish file-format=2 file-path=pointpillars.json ! fakesink
| | [g3dlidarparse](./g3dlidarparse.md) | Parses 3D LiDAR binary frames and attaches custom metadata with point cloud data. Reads raw LiDAR frames (BIN/PCD), applies stride/frame-rate thinning, sets GStreamer timestamps (PTS/duration) for synchronization, and outputs buffers enriched with LidarMeta (points, frame_id, timestamps, stream_id) for downstream fusion, analytics, or visualization.
Example:
gst-launch-1.0 multifilesrc location="lidar/%06d.bin" caps=application/octet-stream ! g3dlidarparse stride=5 frame-rate=5 ! fakesink
| | [g3dinference](./g3dinference.md) | Runs PointPillars 3D object detection on LiDAR point clouds produced by `g3dlidarparse`. It consumes `application/x-lidar` buffers with `LidarMeta`, performs OpenVINO inference, and attaches tensor metadata with flattened 3D detections for downstream processing.
Example:
gst-launch-1.0 multifilesrc location="lidar/%06d.bin" caps=application/octet-stream ! g3dlidarparse ! g3dinference config=pointpillars_ov_config.json device=CPU ! gvametaconvert format=json json-indent=2 ! gvametapublish file-format=2 file-path=pointpillars.json ! fakesink
| | [g3dobjectfuser](./g3dobjectfuser.md) | Spatially associates 2D camera detections with 3D radar or LiDAR detections, performs per-modality object tracking, and emits a single output buffer carrying the cross-modal association as analytics metadata. Consumes a pre-muxed `GstAnalyticsBatchMeta` container from `gvastreammux`, projects 3D boxes into image space using per-camera calibration matrices, and links them to 2D boxes via IoU-based Hungarian association.
Example:
gst-launch-1.0 … gvadetect model=yolo.xml ! mux.sink_0 … g3dlidarparse ! g3dinference config=pointpillars_ov_config.json ! mux.sink_1 gvastreammux name=mux output-mode=container sync-mode=first-pts ! g3dobjectfuser calibration=kitti_calib.json ! gvametaconvert format=json ! gvametapublish file-format=2 file-path=detections.json ! fakesink
| | [g3drender](./g3drender.md) | Renders a LiDAR point cloud — with optional camera streams and cross-modal detection metadata — into a BGR video frame. Supports three modes: `bev` (bird's-eye view with metric grid), `perspective` (synthetic 3D camera), and `cam-proj`. Accepts either a raw `application/x-lidar` stream or a `GstAnalyticsBatchMeta` container.
Example:
gst-launch-1.0 … g3dlidarparse ! g3dinference config=pointpillars_ov_config.json ! mux.sink_1 gvastreammux name=mux ! g3dobjectfuser calibration=kitti_calib.json ! g3drender view-mode=perspective width=1600 height=800 ! videoconvert ! ximagesink
| ## Auxiliary plugins | Element | Description | |------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | [gvaanalytics](./gvaanalytics.md) | Performs video analytics using tripwires and zones. Analyzes object trajectories and attaches watermark metadata for visualization of configured tripwires and zones.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect … ! gvaanalytics config=analytics.json ! gvawatermark ! … OUT
| | [gvaattachroi](./gvaattachroi.md) | Adds user-defined regions of interest to perform inference on (instead of full frame). Example: monitoring road traffic in a city camera feed; splitting large image into smaller pieces, and running inference on each piece (healthcare cell analytics).
Example:
gst-launch-1.0 … ! decodebin3 ! gvaattachroi roi=xtl,ytl,xbr,ybr gvadetect inference-region=1 ! … OUT
| | [gvafpscounter](./gvafpscounter.md) | Measures frames per second across multiple video streams in a single GStreamer process.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect … ! gvafpscounter ! … OUT
| | [gvafpsthrottle](./gvafpsthrottle.md) | Throttles the framerate of video streams by enforcing a maximum frames-per-second (FPS) rate. Useful for rate limiting in pipelines or for testing at specific processing framerates.
Example:
gst-launch-1.0 … ! decodebin3 ! gvafpsthrottle target-fps=10 ! … OUT
| | [gvastreammux](./gvastreammux.md) | Muxes multiple video streams into a single pipeline by assembling PTS-aligned batches across pads, with configurable PTS normalization (`sync-mode`). Each output buffer carries `GstAnalyticsBatchMeta` (`streams[0].index` = source pad index, `n_streams` = batch size).
Example:
gst-launch-1.0 gvastreammux name=mux sync-mode=first-pts ! queue ! gvadetect model=$mDetect device=GPU ! gvafpscounter ! fakesink rtspsrc location=rtsp://host:8554/stream ! … ! mux.sink_0 rtspsrc location=rtsp://host:8555/stream ! … ! mux.sink_1
| | [gvastreamdemux](./gvastreamdemux.md) | Routes a muxed stream back to per-source pads based on `GstAnalyticsBatchMeta.streams[0].index`. Must be used with `gvastreammux`.
Example:
gst-launch-1.0 gvastreammux name=mux ! queue ! gvadetect model=$mDetect device=GPU ! gvastreamdemux name=demux demux.src_0 ! queue ! fakesink demux.src_1 ! queue ! fakesink rtspsrc location=rtsp://host:8554/stream ! … ! mux.sink_0 rtspsrc location=rtsp://host:8555/stream ! … ! mux.sink_1
| | [gvametaaggregate](./gvametaaggregate.md) | Aggregates inference results from multiple pipeline branches.
Example:
gst-launch-1.0 … ! decodebin3 ! tee name=t t. ! queue ! gvametaaggregate name=a ! gvaclassify … ! gvaclassify … ! gvametaconvert … ! gvametapublish … ! fakesink t. ! queue ! gvadetect … ! a.
| | [gvametaconvert](./gvametaconvert.md) | Converts the metadata structure to JSON or raw text formats. Can write output to a file.| | [gvametapublish](./gvametapublish.md) | Publishes the JSON metadata to MQTT or Kafka message brokers or files.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect model=$mDetect device=GPU … ! gvametaconvert format=json … ! gvametapublish … ! … OUT
| | [gvapython](./gvapython.md) | Provides a callback to execute user-defined Python functions on every frame. It is used to augment DL Streamer with user-defined algorithms (e.g. metadata conversion, inference post-processing).
Example:
gst-launch-1.0 … ! gvaclassify ! gvapython module={gvapython.callback_module.classAge_pp} ! … OUT
| | [gvarealsense](./gvarealsense.md) | Provides integration with Intel RealSense cameras, enabling video and depth stream capture for use in GStreamer pipelines. | | [gvawatermark](./gvawatermark.md) | Overlays the metadata on the video frame to visualize the inference results.
Example:
gst-launch-1.0 … ! decodebin3 ! gvadetect … ! gvawatermark ! … | | [gvamotiondetect](./gvamotiondetect.md) | Performs lightweight motion detection on NV12 frames and emits motion ROIs as analytics metadata. Uses VA-API acceleration when VAMemory caps are negotiated, otherwise system-memory path.
Example:
gst-launch-1.0 … ! vah264dec ! gvamotiondetect confirm-frames=2 motion-threshold=0.08 ! gvawatermark ! … | :::{toctree} :maxdepth: 1 :hidden: gvadetect gvaclassify gvainference gvatrack gvaaudiodetect gvaaudiotranscribe gvagenai g3dradarprocess g3dlidarsrc g3dlidarparse g3dinference g3dobjectfuser g3drender gvaanalytics gvaattachroi gvafpscounter gvafpsthrottle gvastreammux gvastreamdemux gvametaaggregate gvametaconvert gvametapublish gvapython gvarealsense gvawatermark gvamotiondetect gstelements :::