Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Horses, dogs, and deers(sp), oh my!

This site may earn commission on affiliate links.
I see that more animal recognitions and associated visualizations are coming in 10.13. Can I raise a practical question at this point: Are we going to do Stonehen..., uh, ... does adding such classifiers to the neural neural networks take up valuable (and potentially limited) tensors that could be used to do more practical things, like keep the car from hitting firetrucks?
 
We have hundreds of mule deer wandering around in our small town. It would be a rare day you don't have to slow down for one (or, usually, more) deer crossing the street. Less often, dogs. Way more common than fire trucks.

But there's no reason not to do both.
 
  • Like
Reactions: Jcl007
It looks like we've run over all the loose dogs in our town, so that none are roaming around on the streets. People should not own dogs and let them wander so they get run over, any more than letting children wander onto the streets.

But speaking of animals, we have bears up here, which would go with the title.
 
  • Like
Reactions: DrGriz
But there's no reason not to do both.
Well that's my question - is there a reason not to do both - specifically, since each classifier takes some tens or hundreds of thousands of "neurons" (also referred to as "tensors") on the Tesla AI chip, where are we on utilization of the processing capacity of the current chips? Do we need to only be adding things at this point that make a difference in autonomous driving?

This assumes, of course, that we would all want to avoid a collision with an an animal, be it a horse, a dog, or a "mule deer," and the distinction between the three (and associated visualization thereof) is a much lower priority than the continued improvement in autonomous driving.
 
Well that's my question - is there a reason not to do both - specifically, since each classifier takes some tens or hundreds of thousands of "neurons" (also referred to as "tensors") on the Tesla AI chip, where are we on utilization of the processing capacity of the current chips? Do we need to only be adding things at this point that make a difference in autonomous driving?

This assumes, of course, that we would all want to avoid a collision with an an animal, be it a horse, a dog, or a "mule deer," and the distinction between the three (and associated visualization thereof) is a much lower priority than the continued improvement in autonomous driving.
Probably easier to just categorize an unknown object as a "box", and something to avoid. If a human sees something on the road that you cannot identify, you just avoid it. It's neat to see the rendering of a dog or cat on the screen, but not really necessary to render everything - like a metal ladder that's fallen off a truck and sitting in the middle of a lane.
 
  • Like
Reactions: Goose66 and DrGriz
I have a spot on my drive home where sometimes the car screams emergency and hits the brakes. The visualization is of a person. But it is actually a power pole at the top of a hill on the side of the road.

I don't think the visualizations are that exact. Something with four legs can be a dog, moose, cow, sawhorse, whatever.
 
I see that more animal recognitions and associated visualizations are coming in 10.13. Can I raise a practical question at this point: Are we going to do Stonehen..., uh, ... does adding such classifiers to the neural neural networks take up valuable (and potentially limited) tensors that could be used to do more practical things, like keep the car from hitting firetrucks?
Avoiding animals like horses, deer, cattle, moose, etc. is as practical as avoiding firetrucks. Dogs, other small animals less so. But having a deer come through the windshield can be a bit of an inconvenience.
 
  • Like
Reactions: DrGriz