PhotonVision
Add object detection training docs
I'm going to try following these instructions and see if I can get it to work :+1:
I will boot up a vm later and try this process from scratch and look for anything I missed.
I was able to follow the instructions and got the model running on an Orange PI. I'll verify tonight if the model is working.
As for the state of this PR - I think this is a huge help for the PV developers. I'm no longer concerned about the bus factor regarding training object detection models. However, I don't think these instructions would be enough for the average team.
I wonder if we can just change the name of this content from "training a custom model" to "converting trained model to RKNN" for now. I think a proper "training custom model" section would require teaching how to set up an environment, installing all of the dependencies, how to label data, export to YOLO formats, training parameters, and only then conversion and uploading.
Yeah broadly agree with @Alextopher -- it's a great set of docs for developers, and as long as we call that out i'd like to get this shipped before it bitrots :)
Sorry for the late response, I've been busy with school and theater. Here is a revised version of the docs that removes the model training part, I will work on a update for the conversion script to just take in a directory instead of a file with paths and clean it up a bit.
Is the scripts folder a good place for the conversion script to go in the PV repo?
Any chance we can get this shoved in for beta (maybe tentatively this weekend?)
Covered by #1723 and #1715. We can still leave it open to document this alternative process, but maybe make it a draft?
