5 ms·
Can someone with more computer vision knowledge explain why we can't teach a machine learning algorithm to detect stationary cars. I have seen other Tesla demo
by linkadin 8y ago
Can someone with more computer vision knowledge explain why we can't teach a machine learning algorithm to detect stationary cars.
I have seen other Tesla demo videos where it clearly detects stationary cars. It seems ridiculous to say that it is a shortcoming of camera when our own two eyes do it.
- Symmetry 8y agoIt's not that it either always detects it or never detects it with video so much as that it doesn't detect it nearly as frequently as a human would. By contrast radar is pretty much always going to see things that aren't stationary.
- linkadin 8y agoBut it would seem to me like that is just a matter of needing more training, or maybe more processing power. At some point with enough training data it will detect a stopped car and react much more quickly than a human would. And even if it's not perfect, all it has to do is beat a human to be worthwhile.
- rayiner 8y agoIt's not so easy, because you don't just have to detect stopped cars, you have to reject things which aren't stopped cars. False positive rejection appears to be something both Tesla and Uber are struggling with. (In the Arizona fatal accident, the Uber vehicle had Uber's own emergency braking system disabled because it was throwing up too many false positives.)