- cross-posted to:
- technology
- [email protected]
- cross-posted to:
- technology
- [email protected]
Repo: https://github.com/ciupava/maptcha_dev
Demo: https://maptcha.crown-shy.com/
I didn’t make this I just wanted to share here before I add it to my weekly urbanism roundup newsletter https://urbanismnow.com/
You must log in or register to comment.
Their html is awful. They need to start over. This is bad.
On Firefox on a tablet, vertical is okay but often there’s nothing outlined in the picture. In landscape the picture covers the top half of the buttons.
I’m legally blind and just tried the demo and it’s god awful.
Edit: spelling
How does one become illegally blind?
LOL, I didn’t even notice that. I guess file a false report with your government. And somehow get it accepted as being true.
“Click on the images where the red outline is clearly marking a building”
Gonna be honest, accessibility for those with altered color vision is gonna need some work.
It shouldn’t be that bad to implement alternate colors.
better yet, make it a shifting line. Should be trivial to spot if the line is shifting through all the chroma and lux values.
I can get behind this one tbh…
Fuck google and sundar the creep
Is he a creep because of the google data privacy stuff or something else?
Correct, Google executives with you behave like Harvey Weinstein with young women actors.
I wish you could say “I’m not sure” because I feel bad getting them wrong 😭
Don’t let perfect be the enemy of good
Easy to say of it would only be used to classify/grade data but because this is a CAPTCHA it somehow has to determine if you answer is “correct” and probably does that by checking if other people agreed with you on the majority of cases and then it gets tricky because one would not only have to take into account if the outline is correct but one thinks that other people would agree with that assessment if you don’t want to get another set of images because the server couldn’t tell of you are a human.
according to the FOSSDEM summary it takes actual human created outlines and mixes it in with the AI stuff. It also takes into account human error so the result after completion is a accuracy score.
Demo doesn’t work in Tor Browser
Are all the demo images supposed to be ‘incorrect’? 'Cause your outlines are all crooked, they look much worse than e.g. Microsofts building outlines and of course proper manualy made ones
It should be (according to the video) but maybe sometimes you get an all bad set?
After like 2 red boxes none of my images had red boxes. Iron fox on android
This could be a great thing for osm!
Does this actually work? Also, what happens if the bots learn to contribute to OSM?
Also, what happens if the bots learn to contribute to OSM?
Never seen that one. That’s good. we need that on Lemmy.
Not even comments. So many posts here are just low effort images. Ive even seen screenshots of headlines posted without a link to the source article. And it got upvotes! Smh
If they make good edits then… yay?
This is dope as hell.
link to the demo https://maptcha.crown-shy.com/
Makes you wonder what people are thinking.
So many rounded corners and buildings with notched taken out of them when they clearly just have a tree over a bit of the building.
I’m pretty sure many of those are “AI” edits or suggestions.
This looks like OpenCV at work.
