I wonder if it would be possible to use a fine-tuned LLaVA 1.6 along with the GPS/phone orientation and reference tape measure etc. in images collected via app.

I wonder if it would be possible to use a fine-tuned LLaVA 1.6 along with the GPS/phone orientation and reference tape measure etc. in images collected via app. I think if you can cover from multiple directions and aggregate the data effectively it could be pretty efficient. Not saying it's easy but I want to believe that a fine tuned open vision model can do a lot. Or just use Claude or GPT 4 vision.