There's some wrinkles: you'd need to do some smoothing to filter out shots of accidents from the day before, say, and also filter out people who might try to grief the system, but it'll work pretty well right out of the box.
You know, that was one of my first thoughts on reading your description. STOP READING MY MIND! ESPECIALLY BEFORE I'VE EVEN HAD THE THOUGHT. ;-)
Seriously, though, this is similar to Google's original PageRank and how it worked nearly "out of the box" on an Internet that wasn't yet affected by Google's relevance. You look at the phrases people use to link to pages, and you get an automatic sense of how the community perceives the linked resource. The more links to a resource, the more interesting it must be. Replace "link" with "geotagged photograph" and "resource" with "location", and voila!
Now, once Google existed, it changed how people linked things. You started getting link spammers and so on, and PageRank had to learn new tricks to sort the wheat from the chaff. I suspect, to prevent SREO (Scenic Route Engine Optimization) from clogging things up with crud, you'll need another layer on this. (Remember Sturgeon's Law: 90% of everything is crud. This is what we're fighting here. The whole point of the Scenic Route score is to find that 10% tht isn't!)
Google's already doing a Street View. Suppose it lets you preview a proposed scenic route in Street View along with some subset of geotagged photos. Then suppose it also lets you tag the photos you see with keywords and maybe a simple thumbs up/down indication. That could serve as an additional check on the "scenic-ness" of a route, as well as provide information on what kind of scenic-ness it offers.
For example, some routes are truly their most beautiful only in certain times of year, such as when trees begin turning color in the fall. The rest of the time, they might be "meh." You could tag the route that way. Others might be interesting if you're into geological formations-my wife, the geologist, loves "road cuts" into the rock-but others might not care so much.
Yep! But it should work long enough to get it working, and then you've got money coming in that you can use to counteract everyone trying to game it. There'll always be people taking shots of their business, and once you filter them, they'll pay others to take shots of their business, and then you build a network of trustworthiness and so on and so on until it becomes sentient!
Ryan, I like your idea! But I don't know if the quantity of photographs in an area is a good metric to determine scenic-ness. In residential areas, there will be all sorts of pictures taken of people inside homes and in back yards and whatnot. I'd actually imagine that would be where the majority of pictures would be taken (which would be interesting to test in itself!).
I would argue that at the very beginning the data would be jumbled with all of those pictures which aren't really scenic. Of course, you could always exclude any pictures with faces in them, then you would be theoretically left with pictures of scenery.
I'd probably argue that at the beginning you would need some human hand to analyze the data of picture counts and assign roads their initial "scenic" number.
Interesting, I hadn't considered that! You could probably look for big swaths of sky and ground-like colours for really-rough "is this outside" metrics. Also, tagging would help some with that too: "tree" and "mountain" are probably outside, "sister" maybe isn't.
Anyway even with these bumps, someone needs to do this, someone needs to make this happen
You could probably use neural network software to sort out the landscape photos too; kickstart it with some known data and then classify photos automatically using that! Photo analysis can be pretty sophisticated these days, I'm sure Google could manage it!
I think if this took off, the GPS should have some sort of button on it so that it could ask "Is this scenic enough for you? Y/N" and when "Y" or "N" is pushed, record the location (and date/time of course) and submit it to the great database in the sky. That would help continually improve this data, as long as you have scenery junkies providing feedback.
I also like the idea of scenery junkies taking the initial feedback of "How much are you willing to be delayed for some pretty scenery?" and answering "I don't care" and getting some really crazy paths to go to the grocery store.
There's also no reason why, down the line, you couldn't classify the scenery. Give me the scenic route with trees! Give me the scenic route with cliffs! Give me the scenic route with neighbours who never close their curtains while changing!
My original response was pretty much that people probably photograph more stupid stuff than they do scenic views, but as that's already been said and the idea is a good one despite the initial hurdles, I'll proceed to the expansion of the idea which immediately occurred to me... which is off-route activities.
Pretty roads are awesome, but for me what makes them interesting is the new experiences you find on the way. That could be as simple as momentary amusement discovering a gas station with a random saddle section (why do they sell saddles, and if you ride and need a saddle why would you think to buy it at a gas station?), or discovering a yummy road food type restaurant, or it could be as complex as yearly festivals and the like. If we're able to tell our GPS to take us on the prettier drive, we should also be able to tell it, "Hey, I want to stop for lunch at a nifty hole in the wall diner with unexpectedly fabulous food, what's the best one for my tastes and budget?," or "Hey, I'm in the mood to attend a little craft show, is there one on my way?" Now, I know that GPS already lists chains and gas stations and the like, but it'd be nice if that were expanded to include little local places and festivals and events, with keyword tagging and a user ratings system to help you find them.
This is an awesome idea, though what immediately popped in my head is how confused travelers in Kyrgyzstan residents will be when iPhone photos from NYC start popping up:
You know, that was one of my first thoughts on reading your description. STOP READING MY MIND! ESPECIALLY BEFORE I'VE EVEN HAD THE THOUGHT. ;-)
Seriously, though, this is similar to Google's original PageRank and how it worked nearly "out of the box" on an Internet that wasn't yet affected by Google's relevance. You look at the phrases people use to link to pages, and you get an automatic sense of how the community perceives the linked resource. The more links to a resource, the more interesting it must be. Replace "link" with "geotagged photograph" and "resource" with "location", and voila!
Now, once Google existed, it changed how people linked things. You started getting link spammers and so on, and PageRank had to learn new tricks to sort the wheat from the chaff. I suspect, to prevent SREO (Scenic Route Engine Optimization) from clogging things up with crud, you'll need another layer on this. (Remember Sturgeon's Law: 90% of everything is crud. This is what we're fighting here. The whole point of the Scenic Route score is to find that 10% tht isn't!)
Google's already doing a Street View. Suppose it lets you preview a proposed scenic route in Street View along with some subset of geotagged photos. Then suppose it also lets you tag the photos you see with keywords and maybe a simple thumbs up/down indication. That could serve as an additional check on the "scenic-ness" of a route, as well as provide information on what kind of scenic-ness it offers.
For example, some routes are truly their most beautiful only in certain times of year, such as when trees begin turning color in the fall. The rest of the time, they might be "meh." You could tag the route that way. Others might be interesting if you're into geological formations-my wife, the geologist, loves "road cuts" into the rock-but others might not care so much.
And so on...
Reply
Reply
I would argue that at the very beginning the data would be jumbled with all of those pictures which aren't really scenic. Of course, you could always exclude any pictures with faces in them, then you would be theoretically left with pictures of scenery.
I'd probably argue that at the beginning you would need some human hand to analyze the data of picture counts and assign roads their initial "scenic" number.
Reply
Anyway even with these bumps, someone needs to do this, someone needs to make this happen
Reply
Reply
Reply
(The comment has been removed)
Reply
I also like the idea of scenery junkies taking the initial feedback of "How much are you willing to be delayed for some pretty scenery?" and answering "I don't care" and getting some really crazy paths to go to the grocery store.
Reply
Reply
Reply
Reply
Reply
Pretty roads are awesome, but for me what makes them interesting is the new experiences you find on the way. That could be as simple as momentary amusement discovering a gas station with a random saddle section (why do they sell saddles, and if you ride and need a saddle why would you think to buy it at a gas station?), or discovering a yummy road food type restaurant, or it could be as complex as yearly festivals and the like. If we're able to tell our GPS to take us on the prettier drive, we should also be able to tell it, "Hey, I want to stop for lunch at a nifty hole in the wall diner with unexpectedly fabulous food, what's the best one for my tastes and budget?," or "Hey, I'm in the mood to attend a little craft show, is there one on my way?" Now, I know that GPS already lists chains and gas stations and the like, but it'd be nice if that were expanded to include little local places and festivals and events, with keyword tagging and a user ratings system to help you find them.
Reply
Reply
http://www.macworld.com/article/134471/2008/07/geotag_errors.html
Reply
Leave a comment