Tuning recommendations guideline

This page contains tuning recommendations for device models that we offer. You can follow them to assure smooth operation of your device.

Jetson Nano

Jetson nano is the weakest in terms of performance from the entire family of jetson products, so it is necessary to pay more attention to tuning.

Due to the complex data processing, we recommend using a maximum of 4 input streams for realtime analyses.

When setting up streams, configure both high and low quality streams. Use the low quality stream for recording and smart recording.
If your cameras output more than 10 frames per second, use the drop frame interval to adjust the frame rate to approximately 10. (e.g. if your camera has 20 FPS, set drop frame interval to 2. So 20/2=10fps)

We do not recommend using a high quality stream for smart recording, especially in connection with notifications. Because most notification services have a limit on the size of uploaded video files. And when using a stream with high quality, this limit is exceeded and notifications are delivered with errors.

When setting up pipelines, configure the AI Detector to operate on every 5th frame.
The min. precision for most classes should be around 40%. Under 20% system may generate more false alarms. Use values over 85% only for objects where you want very high confidence in detections (WARNING: This value will also filter out most true detections).

When setting up alerts, try to avoid objects on scene, as this metric is prone to false detections.

This method can react the fastest to events, but it is also the most prone to interference. So it can generate a lot of notifications.

It is better to use ROI, which can be set up so that disturbing objects are not in it. If you want to achieve even better resistance against false short detections, you can use the "roi occupation" metric, where you also add the minimum time that the object must be in the region for an alert to be created.

If you are not sure what minimum times to set for this metric and a specific stream, then if the system has been running for some time, you can use the integrated Grafana and view the maximum values of the "detection.roi_occupation.5s_val" metric using the "Explore" tool.