>>> 5.6 seconds before impact, it classified her as a vehicle. Then it changed its mind to “other,” then to vehicle again, back to “other,” then to bicycle, then to “other” again, and finally back to bicycle.
This is exactly why I keep saying that autonomous vehicles are not 10 or even 20 years away. More like 50-100 years away if that. Same reason as to why famously a group of researchers in the 60s thought that solving computer recognition of objects would take few months at max, and yet in 2019 our best algorithms still think that a sofa covered in a zebra print is actually a zebra with 99% confidence.
Had a human been actually paying attention to the road, I can bet they would start breaking/swearving as soon as they saw something, even if they weren't immediately 100% certain that it's a human - a computer won't until it's 99%+ certain, which is too risky assumption considering the state of visual recognition of objects.
This is exactly why I keep saying that autonomous vehicles are not 10 or even 20 years away. More like 50-100 years away if that. Same reason as to why famously a group of researchers in the 60s thought that solving computer recognition of objects would take few months at max, and yet in 2019 our best algorithms still think that a sofa covered in a zebra print is actually a zebra with 99% confidence.
Had a human been actually paying attention to the road, I can bet they would start breaking/swearving as soon as they saw something, even if they weren't immediately 100% certain that it's a human - a computer won't until it's 99%+ certain, which is too risky assumption considering the state of visual recognition of objects.