Camera Inputs

Use RTSP for IP cameras that expose a direct RTSP URL.

If the stream does not start, verify the codec first. The dedicated RTSP and RTP inputs are NVDEC-based in the code, so they are only available in builds with the NVIDIA decode path enabled. When using the GStreamer RTSP path, use Convert To Raw GStreamer Pipeline after the simple RTSP setup works if you need to inspect or customize the generated pipeline.

Use RTP when another process sends encoded video directly to a UDP port.

If you need RTP with a GStreamer mux, choose RTP (GStreamer) instead and set Listen Port, Codec, and Mux.

Use GStreamer Pipeline when you need full control over the receive pipeline. Oden sends the pipeline into odenvideosink; you can include ! odenvideosink processing-deadline=0 yourself or let Oden append it.

Use Exclusive Start when several pipelines compete for camera, decoder, or network resources during project load. Lower Exclusive Start Priority values start earlier, and Exclusive Start Duration controls the spacing between starts.

For live sources, keep Sync On Sink disabled unless you need the GStreamer timestamps to control presentation. For prerecorded or buffered sources, enabling Sync On Sink can make playback timing more stable.

You can also use GStreamer when a camera feed must be shared with another process before it enters Oden. On Jetson, use zero-copy IPC when available so the camera frames stay in GPU memory. For example, a camera process can publish frames through nvunixfdsink, and Oden Streamer can receive them through a GStreamer Pipeline input using nvunixfdsrc. On JetPack 5, the nvunixfd plugin is bundled with the Streamer installer. On JetPack 6 and newer, install DeepStream or build voysys/nvunixfd and place the plugin .so in the GStreamer plugin path.

Disable HW Decode for this pipeline because the data is raw GPU memory, not encoded video.

On JetPack 6 and newer, the source build path is:

If the Streamer runs in Docker, the same GStreamer plugin must be available inside the container. Bake it into the image or mount the .so from the host.

ROS and ROS 2 can work for proof-of-concept testing, but they add latency and Oden does not ship built-in ROS capture support. Avoid ROS as the production video path unless the added latency has been accepted for the deployment.

Use SRT when the camera or gateway sends SRT through GStreamer. Set the SRT connection mode and address fields, choose the codec, and start the input. In Advanced, increase Pipeline Latency if the connection is unstable.

Use Advanced when the device needs non-default open behavior:

After the input starts, open Controls to tune camera-specific V4L2 controls such as exposure, brightness, contrast, or menu selections. Use Reset All Settings if the camera controls need to return to their defaults.

Use Rescan Devices after plugging in a new capture device. Use Rescan And Show All Devices only when a device is hidden by the recommended-device filter.

Use Advanced for sensor tuning:

Important GigE settings:

Use USB3 for USB3 Vision devices in internal builds that include USB3 capture. Install UsbDk on Windows if Oden reports that it is missing. Select the camera, configure GenICam settings, and start the input. Enable Allow Slow Cameras only for devices that cannot meet the expected USB3 throughput.

Use Argus for NVIDIA Jetson camera modules. Select the Camera, select the sensor Mode, and start the input. After the camera starts, tune auto exposure, exposure time, analog and digital gain, exposure compensation, AWB mode, white-balance gain, saturation, optical black, antibanding, denoise, and edge enhance as needed. Use Camera Crop before start when the sensor mode supports cropping.

Use Clone Stream when one source needs to be reused in another entity. Select the source in Follow Video.

Enable Focus Region when the clone should expose a smaller focus area for packing or monitoring. Enable Copy To Texture when you need independent filtering or hard crop on the clone. Without Copy To Texture, the clone uses the original source texture. Use Hard Crop only when the clone should produce a physically cropped texture instead of only reporting a focus region.

Use Display Capture to capture a monitor. On Windows, Cursor controls whether the pointer is included and Monitor selects the display. On Linux, Oden uses an X11 GStreamer capture pipeline.

Use Test Source to verify layout, packing, encoding, and network output before connecting real cameras. Set Resolution and Pattern. Some static patterns can trigger drop detection, so disable drop-sensitive checks when intentionally using a static pattern.

Use Encoder Test when you need a harder-to-encode generated source. It renders a colorful noise-based sphere and exposes settings such as noise amount, noise frequency, and animation speed. The extra detail is useful when checking encoder quality, bitrate control, and packet-loss behavior.

Use Image for static visual checks. Select a .png, .jpg, or .jpeg file. Alpha is supported for formats that carry transparency.

Use the FFmpeg-backed inputs for prerecorded files, HLS playlists, RTMP streams, or custom FFmpeg input strings. Set URL to the file path or stream URL, set Input Format when auto-detection is not enough, and choose hardware acceleration only when the deployed GPU and driver have been validated. Enable Live Source for sources that should be treated as realtime. Enable Loop for prerecorded files used as repeatable tests. Enable Auto Start when the source should start as soon as the project opens.

Use NDI for network NDI sources. Select the source and set Quality.

Use Spout on Windows for GPU texture sharing from another application. Select Source, then set Use Alpha Channel and Log Spout Messages only when needed. Open Info to inspect whether the transfer path stays on the GPU or goes through CPU memory. Spout is a same-machine Windows GPU sharing path, so validate that the sending application and Oden run on the same GPU when avoiding CPU copies matters.