Image Annotation

So, I stumbled upon the BeamNG.research page and was looking at some of the different videos about the AI system. I understand the concept of the Lidar sensors (super cool btw), but I'm unfamiliar with the "Pixel Perfect Image Annotation". Is this just live image/object recognition, or is it something else?

 

Oh wow. So you guys are training the AI system just about the same way real autonomous vehicles are taught to navigate. That's really amazing!

 

My understanding is that what is in .Research is not necessarily in .Drive, for example how AI is taught to see can be bit different as in game there would be easier and much more efficient ways of handling that aspect, but for .Research which aims to provide platform for autonomous vehicle testing, different approach is needed.

Naturally some aspects will transfer between the products, but I would think that for .Drive full simulated machine vision would be a total overkill and only purpose of such would be development of machine vision simulation for .Research project.

.Drive user community is a strength there that we can test far larger array of possibilities than any dedicated testing team can, which might be beneficial at some stage of developing machine vision simulation, but I would think that considering needed resources and computing power, such might be not best suited for game purposes.

Then again, they do simulate skeleton vehicle consisting of beam and nodes at 2000Hz when you could get just fine crashing with simpler methods, so there might be little bit of passion to mess with boring logic and numbers

 

Are we ever going to be able to download the files? https://projects.beamng.com/projects/research/files

 

Ah, that makes sense. Thanks