On Thu, Jun 17, 2021 at 7:25 AM Dennis Lee Bieber <wlfr...@ix.netcom.com> wrote: > > On Thu, 17 Jun 2021 06:07:59 +1000, Chris Angelico <ros...@gmail.com> > declaimed the following: > > > >How well can you define the things you're looking for? > > > >https://xkcd.com/1425/ > > Non sequitur comment -- Sounds like a potential usage of a BeagleBone > AI with the TI Deep Learning modules... though I don't think they've > documented how to go from the example learning images to teaching it to > recognize birds <G>
Yeah. We have the tools for high-grade AI now, but.... ultimately, teaching an AI to recognize birds requires someone spending quite a few hours going through photos of birds and drawing boxes around them. "That's a bird. That's a bird. That's a bird. That's not a bird, it's a plane. That's a bird. That's a bird. That's not a bird, it's Superman." > Won't help the OP unless it can be taught to recognize vehicles and > somehow provide coordinates to let the OP "blank out" said vehicles. > > Edge detection might provide some hints -- but simple "same color > pixels" won't -- as the roads may have different colors (concrete vs > asphalt vs gravel vs dirt -- and then you get asphalt/tar patches applied > to pot-holes and cracks in concrete). Also, as in that photo, many of the > vehicles are near enough to the road color that any range used to ensure > picking up variations in the road will pick up the vehicles. > Yup. Image processing is hard. Having seen what people can do with various tools, I can only conclude that (a) the tools are magic, and (b) the job is STILL a big one, even with magical tools. ChrisA -- https://mail.python.org/mailman/listinfo/python-list
