Fusion hardware requirements

in our lab, we are interested in using Fusion for multi-camera body tracking; however, from the doc at Fusion - Stereolabs, the hardware requirements for the two proposed workflows local and network are not clear.

We are planning to use 4 cameras.

  1. Do we need specific hardware for computing in both cases? In the network case, zed boxes that are essentially Nvidia Jetson running Ubuntu 20 are proposed. We are working in a ROS2 environment so that it might be inconvenient. Can we use any PC with GPU running Ubuntu >= 22?

  2. For both cases, an “AI server datacenter” is required. Can we have some info on what the requirements would be for it?

Hi @gionic,

Welcome to the forums!

Any hardware that supports the ZED SDK can be used for a Fusion configuration, more info on that here: Stereolabs Docs: API Reference, Tutorials, and Integration

  1. Indeed, if your setup consists of having multiple Ubuntu PCs with GPUs in the network configuration, this is compatible with our solution. The latest ZED SDK currently supports Ubuntu versions 20 and 22 FYI.
  2. As I’ve mentioned before, any PC which supports the ZED SDK can serve as the “AI server datacenter”. The diagram you’ve linked mentions that in the “local” workflow, the server must have more important CPU and GPU resources as it computes our algorithms for all 4 cameras. For your system of 4 cameras, a high end CPU and a PC with an RTX4090 is recommended to have real-time performance. In the “network” config, our algorithms are distributed to each worker, so the central server requires less compute.

Hope this helps answer your questions