It seems that you're using an outdated browser. Some things may not work as they should (or don't work at all).
We suggest you upgrade newer and better browser like: Chrome, Firefox, Internet Explorer or Opera

×
Community
General discussionAny idea how to solve the google reCaptcha street signs puzzles?(40 posts)(40 posts)
(40 posts)
Pages:
1
2
3
This is my favourite topic
Posted August 06, 2018
The W3C has had a public notice posted about "Inaccessibility of CAPTCHA" since at least 2005.

About a month ago they started to revise and update a working draft for a new version of "Inaccessibility of CAPTCHA". They're using GitHub for comments and changes, but also have an email address in case that isn't feasible.

Of particular note in that document is section 2 for "Security effectiveness", or rather the ineffectiveness of CAPTCHA.

The W3C also has "Web Accessibility Initiative" (WAI) on their website, with news about progress along those lines.

However, the W3C can only make recommendations. Software developers have to figure out the implementation for each circumstance encountered based on their available resources in the moment. Unfortunately, they're only human.
Post edited August 06, 2018 by thomq
Posted August 06, 2018
Annoying AF. Finding an alternative to Google SE, Youtube, and Chrome again.
Posted August 06, 2018
Madness;
https://www.youtube.com/watch?v=7WhrMMsmJRY
https://www.youtube.com/watch?v=5BggRMRzrlo
https://www.youtube.com/watch?v=8ejSwXR-HBs
Posted August 09, 2018
Nothing better than when recaptcha asked me to identify road signs in a picture that only included chinese characters. Probably. Could be korean or japanese too, I don't speak any eastern language so they all look the same to me.

I had no idea if the signs in the picture were road names, ads, tourist attraction info or whatever else.
Posted August 09, 2018
I usually only select 3 boxes which contain the most part of the signs.
Posted August 10, 2018
I've never selected the poles in any captchas and got 2 or 3 recaptchas at most.

From what I know, the captchas are used to train Google's algorithms for Google Maps, Street View or whatever things they're using recognition software on.

Signed.
Posted August 10, 2018
Why don't you skip them?
Posted August 10, 2018
Last time I tried to log into Humble, it took me at least 8 iterations, and the stupid image loads kept purposefully slowing down each time. And that was just to log in. THEN I needed to do it again to get to my downloads page. AND THEN I needed to do it again to go to the "email me unclaimed keys" page (for another email account that gets them). And each time it took a few tries.

It might have been because I was using my linux box at the time. I think it reads your stuff and gives more to linux?
Posted August 10, 2018
Don't take it personally when Google says you can't tell a street sign from a helicopter. You can. But at that point you're not being presented a test to prove you're a human. You're being presented unclassified images to help train their image recognition algorithms. Years ago, when they were teaching their software to read, it was digitized books, but as of late their "tests" are geared towards their self-driving car project, and also mapping the world via street-view images instead of third-party cartographic data.

Here's how it works:

- They have some image that probably contains a street sign according to some preliminary (automatic?) analysis, but they're not really sure.
- They want to make sure, so they start showing the image to human beings. That's us (actually working for free for them).
- Human beings click squares, but at that point Google doesn't know yet if we're clicking street signs or turds on the road. And they don't know if you're a human or a program clicking randomly.
- What do they do? Nothing, just write down what you've clicked and show you another image without giving any feedback, which is what irks people.
- Eventually they notice most human beings click the same squares on that image, so now they can be sure that street signs are there. Now the image is correctly classified, and they can actually use it as a recaptcha test and, more importantly for them, to train their self-driving car.
- Eventually you get one of the already classified image, and that's the one you have to get right. The previous ones didn't matter at all to pass the test. Don't feel stupid.

I don't know the exact details of how tests are interspersed with unpaid training work, but certainly there must be some randomness in the process. Maybe you have to pass 1-2 tests, maybe you get more than your share of unclassified images if you fail an actual test, maybe they just got a huge batch of images that look like store fronts and they're in a hurry to classify them so they'll show 5 of them to everybody... I don't know, but anyway that would explain why some times it takes 2-3 tries and others it's 9-10 tries and a smashed mouse.
Posted August 10, 2018
avatar
nepundo: Don't take it personally when Google says you can't tell a street sign from a helicopter. You can. But at that point you're not being presented a test to prove you're a human. You're being presented unclassified images to help train their image recognition algorithms. Years ago, when they were teaching their software to read, it was digitized books, but as of late their "tests" are geared towards their self-driving car project, and also mapping the world via street-view images instead of third-party cartographic data.

Here's how it works:

- They have some image that probably contains a street sign according to some preliminary (automatic?) analysis, but they're not really sure.
- They want to make sure, so they start showing the image to human beings. That's us (actually working for free for them).
- Human beings click squares, but at that point Google doesn't know yet if we're clicking street signs or turds on the road. And they don't know if you're a human or a program clicking randomly.
- What do they do? Nothing, just write down what you've clicked and show you another image without giving any feedback, which is what irks people.
- Eventually they notice most human beings click the same squares on that image, so now they can be sure that street signs are there. Now the image is correctly classified, and they can actually use it as a recaptcha test and, more importantly for them, to train their self-driving car.
- Eventually you get one of the already classified image, and that's the one you have to get right. The previous ones didn't matter at all to pass the test. Don't feel stupid.

I don't know the exact details of how tests are interspersed with unpaid training work, but certainly there must be some randomness in the process. Maybe you have to pass 1-2 tests, maybe you get more than your share of unclassified images if you fail an actual test, maybe they just got a huge batch of images that look like store fronts and they're in a hurry to classify them so they'll show 5 of them to everybody... I don't know, but anyway that would explain why some times it takes 2-3 tries and others it's 9-10 tries and a smashed mouse.
Thanks for the informative explanation!

Just to add my non-scientific experiences:
- Clicking on the scenes with signs is generally just a waste of time, as it very rarely gets you past the captcha.
- I get the feeling that the number of screens they force you to click through is at least partly tied to how much the site wants, or even cares, if you actually enter the site.
- Scenes with cars, roads, buses or store fronts are much more likely to be successfully passed. These contain three images that you should click on (and any additional correct images that reappear afterwards).
- I sometimes wonder why some sites don't just go the whole hog, and when you click on the "I'm not a robot" button, a sign pops-up that says "F@#K OFF!" (possibly with pointlessly clickable segments!).
Pages:
1
2
3
This is my favourite topic
Community
General discussionAny idea how to solve the google reCaptcha street signs puzzles?(40 posts)(40 posts)
(40 posts)