[Question] Best model for indoor usage

[aiotko]

We're trying to use FastMOT for processing video from surveillance cameras installed in the office.

The default YoloV4 model works very well at the same time it requires a powerful server for dealing with tens video streams in parallel.

Pre-trained YoloV4Tiny from Darknet is much faster and not so demanding but doesn't work well especially if infrared surveillance subsystem is used.

Maybe you could advice which model will be "balanced" for indoor usage as well as which datasets might be used for training YoloV4Tiny (or another model)?