Performance Tuning

When using the Gravio Hub for inference computer vision detection, please consider the following:

When an inference is selected as a “layer as a soft sensor” on a Gravio Hub, only one model should be used per Gravio Hub.
Depending on the complexity of the model file, the load on memory and CPU may vary.
For a typical TensorFlow model, we recommend that you use no more than two cameras connected and that you run the model at intervals of about one minute.
Make sure to run the camera images at least 15-20 seconds apart.
In the case of the Lite TensorFlow model, it is recommended that no more than two cameras are connected at any given time, and that the recognition runs at intervals of 30 seconds.
To keep the Gravio Hub at optimum performance, log into the Gravio Hub via SSH and use the top command to see the average load. For ideal performance, please keep the load average below 2.

If you plan to operate the HubKit at a rate exceeding this recommendation, please install the HubKit on a dedicated PC with higher processing power.

The disk capacity of Gravio Hub is 16GB, including the OS.
The available space for operation is approximately 6GB.
If you have enabled image output during inference, please be careful not to exceed this capacity.

The Gravio Hub supports the Coral USB Accelerator by Google if you require extra power for your Gravio Hub. (Not included in the package).