Running Sentis on the cloud

Wondering if anyone tried to use Sentis on the cloud? I know it’s meant to be used on-device, to get rid on network latency, but there are use cases where inference latency can be longer than network latency. In that case, it would make sense to run a model on the cloud/edge to provide a faster response time.
A few examples: models requiring GPU, Unity running on slow CPU device, model is too large for on-device, etc.
So running a Unity project with Sentis, and then making requests from the device, could make its use more broad.
Any thoughts from the community on this?

1 Like

Just to clarify. You want to run a game on a mobile or VR headset, for example, but have the model running on a PC with a good GPU?

Yes, this can be a good idea. For example that is how some VR headsets work. One way to do this would be to set Unity as a server and send web requests. You’d probably find more information in the networking sections of the forums.

Perhaps this is something we should write a tutorial on if anyone else is interested in this?

Yes pretty much. Have the game run locally, but make requests to a remote Unity server using Sentis loaded with a model.
Now, I know this could be done without Sentis, just loading a model in a Python environment, but I feel Sentis can make this more accessible to game developers.
What we would need for this is an example of a request made from the client, to the server. I’m not sure how we would send the information to the Sentis server, get it processed, and have it send back the result to the client.

1 Like