What's your Google captcha algorithm when confronted with odd shapes?

I run FF Google searches on a proxy for the obvious privacy-from-google
reasons which used to work 100% of the time without captcha challenges.

Now, it works only about 80% of the time without captcha challenges.
And, increasingly with FF, I have to suffer well more than 3 challenges.

When I get the Google search captcha challenge, sometimes it asks for fire
hydrants, which are pretty easy so it usually works within three tries if
the next two tries are also simple things like chimneys and or cars/buses.

But often it's odd shaped things like bicycles, motorcycles, crosswalks,
stairs and even streetlights seem to defy my ability to satisfy captcha.

I give up after three tries, which means half the time now when I'm
confronted with the captcha, I end up giving up and going to DDG instead.

But my main question here is that I happen to be human and yet Google can't
figure that out in three tries (which is all I have the patience for).

What do YOU do when confronted with odd shapes like motorcycles & bicycles?
Do you click only on the main three squares? Or do you hit everything?

Same with the crosswalks which may protrude into a corner of a near square.

Do you click on every mirror and helmet of the bike riders?
Do you click on every square touched by a crosswalk?
Do you click on every square with streetlights when they're distant?

Whatever method I'm using, isn't working. So I'm not acting like a human.

How do you act like a human when you are using FF with a proxy and as a
result you are suffering innumerable Google search captcha challenges?

On 5/20/2023 12:10 AM, vader wrote:
> What's your Google captcha algorithm when confronted with odd shapes?
>
> How do you act like a human when you are using FF with a proxy and as a
> result you are suffering innumerable Google search captcha challenges?

Part of acting human, is to appear fallible. Taking too long
to decide. Ticking and unticking a box because you aren't sure.

If Google wants, they can just make you endure three trials,
no matter how you answer :-) There is that.

Getting the four boxes right, in fifteen milliseconds, is a bad thing. OK ?

Paul

vader wrote:
[..snip..]
> I give up after three tries, which means half the time now when I'm
> confronted with the captcha, I end up giving up and going to DDG instead.

Think about that.

[..snip..]
> Whatever method I'm using, isn't working. So I'm not acting like a human.
>
> How do you act like a human when you are using FF with a proxy and as a
> result you are suffering innumerable Google search captcha challenges?

Just don't use Google (directly).

In alt.comp.software.firefox vader <darthvader@victory.net> wrote:
> What's your Google captcha algorithm when confronted with odd shapes?
>
> I run FF Google searches on a proxy for the obvious privacy-from-google
> reasons which used to work 100% of the time without captcha challenges.

I don't use Google or a proxy, but, probably due to my unusual
Firefox configuration, websites like Ebay and PayPal like to force
me through the same thing from time to time, also using Google
Captchas. Sometimes the Ebay ones are just endless, I think
something is broken there which causes it to repeat captchas
forever. Online shopping really makes me dream of physical stores
sometimes.

> What do YOU do when confronted with odd shapes like motorcycles & bicycles?
> Do you click only on the main three squares? Or do you hit everything?

My rule is to click on any square with a bit of the described
object (including attached parts) visible in any part of that
square. That's probably harder for a computer to do, if not
necessarily a normal human reaction to the prompt.

But for traffic lights I only pick squares where the lights
themselves are visible, not the poles.

The one that gets me though are the stairs - should I include the
hand rail as part of the stairs, or just the steps themselves? The
latter seems to work, but the question still haunts me.

> Same with the crosswalks which may protrude into a corner of a near square.
>
> Do you click on every mirror and helmet of the bike riders?
> Do you click on every square touched by a crosswalk?
> Do you click on every square with streetlights when they're distant?

Yes to all of the above, and although it does often hit me with two
captchas in a row, I usually get through (except those endless ones
on Ebay where I think it's a different problem because the whole
page reloads).

The audio captcha for blind people is another option, if you have
audio working in Firefox (I don't anymore because I don't use
PulseAudio on Linux and run Mozilla's prebuilt executables which
are compiled without ALSA support, but if you're using Windows
then you probably do).

--
__ __
#_ < |\| |< _#

On Fri, 19 May 2023 23:10:00 -0500, vader wrote:

> What's your Google captcha algorithm when confronted with odd shapes?

I'm still in the process of figuring out the principles behind the current
Google re:captcha v3 picture queries. One error source I noticed is, that
sometimes the u-block browser extension blocks relevant traffic, leading
to a high failure rate. Instead of constantly checking all filter lists
I finally settled for whitelisting gstatic.com.

Apart from this, there seem to be 3 different basic query types:

The most simple one shows completely different scenes in each image tile;
and the selection will just stick (without replacement), whenever a tile
is selected. The solution requires sometimes 3, sometimes 4 tiles to be
selected. (I never encountered more or less.) One of these correct tiles
usually differs a lot from the other ones in general appearance.

The second type is somewhat similar to the first, but each selection will
immediately be replaced by an alternate image. This re:captcha seems to
always have 3 tiles to be selected in the first round and at least one
tile in the second selection round. Because only tiles selected in the
previous round are renewed, only these have to be considered in each
subsequent round. Although the test usually ends after selecting only
one tile in the second round, there may be up to 3 tiles that need to be
selected in the second (and more subsequent) rounds. The maximum number
of rounds seems to be virtually infinite. But it seldom exceeds 2 or 3.
So, just be patient.

The most complex re:captcha type shows one image spanning all tiles. This
one requires the selection of all traffic lights, busses, motor cycles,
bicycles, cars, pedestrian crossings and so on. The selected tiles stick
on selection and can be unticked. The button shows the title SKIP (when
no tile is selected) or NEXT (when at least 1 tile is selected) in at
least the first round and the title SKIP (no selection) or VERIFY (at
least one selected tile) in the last round. The number of rounds seem
to depend on a basic "complexity" default setting issued from the target
website, randomness, and a "user is likely human" factor derived from
selection speed, correctness in details and maybe even tile selection
order. VERIFY will (in my experience) /usually/ appear between the third
to fifth round. Sometimes earlier, sometimes later, though. Only the last
round will show, whether the error rate in this and earlier selection
round(s) was too high. That said, the VERIFY round seems to have the most
impact on the decision. Therefore, it is necessary to get this one as
correct as possible.

Major problem with re:captcha type 3 is, that there is no fixed rule,
whether small details have to be included in a selection. Sometimes
(for instance) the mirror of a motor bike is considered part of the
bike, and sometimes not. Sometimes, the background plates of traffic
lights have to be included, sometimes not. In my experience, it is
best to look at the image from further away from the screen, after
all clearly correct tiles are selected. If the remaining small details
are still visible and clearly belong to the object(s) in question, the
tile in question rather needs selection. Otherwise rather not.

The ones riding motor bikes and bicycles, the poles of traffic lights,
the empty area between pedestrian crossing stripes (or accompanying
transversal stop stripes for the cars) and any other "transcendentally
belonging" parts of the selection objects never seem to be part of the
correct solution. The same goes for pedestrian crossing traffic signs
and the like. (Which without the stripes visible on the road obviously
do not count as pedestrian crossing.) Clearly visible bicycles on traffic
signs, OTOH, seem to be /sometimes/ part of the accepted solution. But
more often, they are not. Shadows of objects and /small/ parts of the
objects lost in shadow (tire ends below busses come to mind) usually
are not to be selected.

Another problem with the third type is, that the correctness depends on
correct prior classification. Sometimes small details (for instance
/additional/ pedestrian crossings in the far background of the image)
have to be selected, as well. But sometimes, these details seem to have
been overlooked by the original classification team. (And probably the
majority of the "human" user base, as well.)

Therefore, a certain amount of luck is necessary to get the third type
right. Achieving at least near-correctness seems to be necessary, though.
Because in my experience, near-correctness increases the likelihood the
get the selection types 1 or 2 after a failed test. (Instead of another
type 3...)

All that said: Google surely manages to inflict self-doubt about one's
own humanness with these captchas... ;-)

HTH.
BeAr

F-Up set to acf.
(Seems to me the one with the closest link of the 3 to the topic.)
--
===========================================================================
= What do you mean with: "Perfection is always an illusion"? =
===============================================================--(Oops!)===

vader wrote:

> What's your Google captcha algorithm when confronted with odd shapes?

Perhaps "go to specsavers"?

apologies if that joke doesn't travel well ...

On 20/05/2023 05:10, vader wrote:

> How do you act like a human when you are using FF with a proxy and as a
> result you are suffering innumerable Google search captcha challenges?

You can switch to the option of speech recognition, but it's not always
clear what is being said!

--

Jeff

On 5/20/23 00:10, this is what vader wrote:
> What's your Google captcha algorithm when confronted with odd shapes?
>
> I run FF Google searches on a proxy for the obvious privacy-from-google
> reasons which used to work 100% of the time without captcha challenges.
>
> Now, it works only about 80% of the time without captcha challenges.
> And, increasingly with FF, I have to suffer well more than 3 challenges.
>
> When I get the Google search captcha challenge, sometimes it asks for fire
> hydrants, which are pretty easy so it usually works within three tries if
> the next two tries are also simple things like chimneys and or cars/buses.
>
> But often it's odd shaped things like bicycles, motorcycles, crosswalks,
> stairs and even streetlights seem to defy my ability to satisfy captcha.
>
> I give up after three tries, which means half the time now when I'm
> confronted with the captcha, I end up giving up and going to DDG instead.
>
> But my main question here is that I happen to be human and yet Google can't
> figure that out in three tries (which is all I have the patience for).
>
> What do YOU do when confronted with odd shapes like motorcycles & bicycles?
> Do you click only on the main three squares? Or do you hit everything?
>
> Same with the crosswalks which may protrude into a corner of a near square.
>
> Do you click on every mirror and helmet of the bike riders?
> Do you click on every square touched by a crosswalk?
> Do you click on every square with streetlights when they're distant?
>
> Whatever method I'm using, isn't working. So I'm not acting like a human.
>
> How do you act like a human when you are using FF with a proxy and as a
> result you are suffering innumerable Google search captcha challenges?
Don't you have an option to request another set of images? I know you do with the 4 jumbled letter guesses.

I like the captcha that has one large image broken into 9, rather than 9 separate ones. But still, it's always a pot
luck of chance.
--
Linux Mint 21.1 Cinnamon 5.6.8
Al

On 20-05-2023 12:48 Paul <nospam@needed.invalid> wrote:

> Part of acting human, is to appear fallible. Taking too long
> to decide. Ticking and unticking a box because you aren't sure.

I've read that the automated engines hit the boxes at the same interval, so
there's merit to hitting the boxes in random intervals.

But what's to stop the bots from randomizing their clicks too?

> If Google wants, they can just make you endure three trials,
> no matter how you answer :-) There is that.

Speaking of three, I've noticed many times there are only three of each
item. Have you noticed that or is it just my imagination?

On 2023-05-20, Computer Nerd Kev <not@telling.you.invalid> wrote:
> I don't use Google or a proxy, but, probably due to my unusual
> Firefox configuration, websites like Ebay and PayPal like to force
> me through the same thing from time to time, also using Google
> Captchas.

I also see endless captchas whenever I use Opera or Brave browsers in their
VPN or TOR modes. When I switch directly to Firefox, they stop altogether.

For some reason, Google thinks (real) humans don't use IP obfuscation.

> Sometimes the Ebay ones are just endless, I think
> something is broken there which causes it to repeat captchas
> forever. Online shopping really makes me dream of physical stores
> sometimes.

I think they may make them endless on purpose because they don't want you
to use IP address obfuscation. They force you to use your real IP address.

They figure only (real) humans have a (real) IP address I guess.

>> What do YOU do when confronted with odd shapes like motorcycles & bicycles?
>> Do you click only on the main three squares? Or do you hit everything?
>
> My rule is to click on any square with a bit of the described
> object (including attached parts) visible in any part of that
> square. That's probably harder for a computer to do, if not
> necessarily a normal human reaction to the prompt.

I've always wondered about that. What if a crosswalk protrudes only an
eighth inch into the next square? Would a (real) human click on it?

How does a (real) human handle it when the bicycle has a sixteenth of an
inch of the wheel in the adjacent square? I need to act like a (real) human
so that's kind of useful to know what you (presumably a real human) does.

What do you do, if you want to look like a (real) human to Google captcha
when the fire street signs have a pole that protrudes into another square?

> But for traffic lights I only pick squares where the lights
> themselves are visible, not the poles.

There are sometimes things which could be traffic lights but they're not
facing the user. How does a (real) human handle that? I need to know.
> The one that gets me though are the stairs - should I include the
> hand rail as part of the stairs, or just the steps themselves? The
> latter seems to work, but the question still haunts me.

Yup. Is the handrail part of the stairs? And what if the corner of a step
protrudes into the next square. How do you (real) human's handle that one?

>> Same with the crosswalks which may protrude into a corner of a near square.
>>
>> Do you click on every mirror and helmet of the bike riders?
>> Do you click on every square touched by a crosswalk?
>> Do you click on every square with streetlights when they're distant?
>
> Yes to all of the above, and although it does often hit me with two
> captchas in a row, I usually get through (except those endless ones
> on Ebay where I think it's a different problem because the whole
> page reloads).

I too wondered about the helmet of the rider. Does a (real) human consider
the helmet to be part of the bike? Now I know the answer is yes.

Thank you for training me how to be a (real) human because Google thinks
I've got it all wrong and that I'm just learning how to be a (real) human.

> The audio captcha for blind people is another option, if you have
> audio working in Firefox (I don't anymore because I don't use
> PulseAudio on Linux and run Mozilla's prebuilt executables which
> are compiled without ALSA support, but if you're using Windows
> then you probably do).

I didn't think of the audio captcha myself. Don't think I've seen it.
Do the (real) humans often switch to that? If so, that's what I want.

We all need to train ourselves so we appear to be (real) humans to google.

On Sat, 20 May 2023 15:05:39 -0000 (UTC), Incubus wrote:

Disclaimer: The following detailed suggestions may be wrong. ;-)

> I've always wondered about that. What if a crosswalk protrudes only an
> eighth inch into the next square? Would a (real) human click on it?
>
> How does a (real) human handle it when the bicycle has a sixteenth of an
> inch of the wheel in the adjacent square? I need to act like a (real) human
> so that's kind of useful to know what you (presumably a real human) does.

In my experience, the original captcha training team must have used monitors
showing few details. Therefore my recommendation (in a different post) was
to ignore small parts invisible from a larger (maybe double of the normal)
viewing distance, but to include every still recognizable details.

> There are sometimes things which could be traffic lights but they're not
> facing the user. How does a (real) human handle that? I need to know.

If (from position and look) they seem to be similar to the ones facing
up front, then include. Otherwise not.

> Is the handrail part of the stairs?

In my experience, it should /not/ be included for a "correct" resolve.

> And what if the corner of a step protrudes into the next square.

Like above: Include if still visible from greater distance.

Apart from this: If the steps (or whatever object is looked for) is visible
/through/ an obfuscation (like step handrails, a fence and the like), it has
to be included. But never, when it is completely overlaid by another object.
I.e.: No "projecting" of an object boundary into concealed space.

> I too wondered about the helmet of the rider. Does a (real) human consider
> the helmet to be part of the bike? Now I know the answer is yes.

In my experience, neither the biker nor the helmet are part of the "correct"
solution. And the handlebar and any extension on it only, if well defined
and clearly visible. Today I had a bike with well defined mirror in one
tile and a shady end of the handlebar with break in another (both adjacent
to the bike shown clearly in other tiles). The accepted solution was, to
select the mirror, but not the brake. - Can't say, whether also selecting
the brake would /also/ have been accepted, tough...

> We all need to train ourselves so we appear to be (real) humans to google.

It really is a mess! :-(

BeAr
F-Up set to acf.
--
===========================================================================
= What do you mean with: "Perfection is always an illusion"? =
===============================================================--(Oops!)===

In alt.comp.software.firefox Incubus <u9536612@gmail.com> wrote:
> On 2023-05-20, Computer Nerd Kev <not@telling.you.invalid> wrote:
>
>>> What do YOU do when confronted with odd shapes like motorcycles & bicycles?
>>> Do you click only on the main three squares? Or do you hit everything?
>>
>> My rule is to click on any square with a bit of the described
>> object (including attached parts) visible in any part of that
>> square. That's probably harder for a computer to do, if not
>> necessarily a normal human reaction to the prompt.
>
> I've always wondered about that. What if a crosswalk protrudes only an
> eighth inch into the next square? Would a (real) human click on it?
>
> How does a (real) human handle it when the bicycle has a sixteenth of an
> inch of the wheel in the adjacent square? I need to act like a (real) human
> so that's kind of useful to know what you (presumably a real human) does.

Well I do get two captchas in a row often so I'm not sure if it's
best, but yes I do select any box with any tiny bit of the object
protruding into it.

> What do you do, if you want to look like a (real) human to Google captcha
> when the fire street signs have a pole that protrudes into another square?

I don't think I select the pole because that's not shown in the
example image that they put at the top of the street sign ones.

>> But for traffic lights I only pick squares where the lights
>> themselves are visible, not the poles.
>
> There are sometimes things which could be traffic lights but they're not
> facing the user. How does a (real) human handle that? I need to know.

I think most people would select them, but it's a guess.

>> The one that gets me though are the stairs - should I include the
>> hand rail as part of the stairs, or just the steps themselves? The
>> latter seems to work, but the question still haunts me.
>
> Yup. Is the handrail part of the stairs? And what if the corner of a step
> protrudes into the next square. How do you (real) human's handle that one?

I sometimes select a lot of stair corners and I've been a little
surprised sometimes that it did accept that method even though it
meant a lot of squares were selected.

>>> Same with the crosswalks which may protrude into a corner of a near square.
>>>
>>> Do you click on every mirror and helmet of the bike riders?
>>> Do you click on every square touched by a crosswalk?
>>> Do you click on every square with streetlights when they're distant?
>>
>> Yes to all of the above, and although it does often hit me with two
>> captchas in a row, I usually get through (except those endless ones
>> on Ebay where I think it's a different problem because the whole
>> page reloads).
>
> I too wondered about the helmet of the rider. Does a (real) human consider
> the helmet to be part of the bike? Now I know the answer is yes.

Well actually I usually get parked bikes (and scooters), I don't
really remember one with a rider on it, and I certainly haven't had
enough of them to be sure of whether to include the helmet on a
rider or not.

>> The audio captcha for blind people is another option, if you have
>> audio working in Firefox (I don't anymore because I don't use
>> PulseAudio on Linux and run Mozilla's prebuilt executables which
>> are compiled without ALSA support, but if you're using Windows
>> then you probably do).
>
> I didn't think of the audio captcha myself. Don't think I've seen it.
> Do the (real) humans often switch to that? If so, that's what I want.

It worked reliably for me for a time before I switched to using
Mozilla's executables and lost audio from Firefox. Sometimes you
have to load a different one because it's just noise, but usually
I think it was quicker.

--
__ __
#_ < |\| |< _#

In article <u49h5r$37kkg$1@novabbs.org>, vader wrote...
>
> What's your Google captcha algorithm when confronted with odd shapes?
>
> ...

If we are taking it on trust that only human intelligence can solve one of
these "spot the hydrant" (etc) puzzles, does it not follow that there must be a
rooom full of humans somewhere setting them?

--

Phil, London

On 2023-05-20 08:43, B. R. 'BeAr' Ederson wrote:

....

> F-Up set to acf.
> (Seems to me the one with the closest link of the 3 to the topic.)

Please don't. There are always people that don't read the whatever group
you set follow up to. I don't read alt.comp.freeware, for instance.

--
Cheers, Carlos.

On 2023-05-21 13:05, Philip Herlihy wrote:
> In article <u49h5r$37kkg$1@novabbs.org>, vader wrote...
>>
>> What's your Google captcha algorithm when confronted with odd shapes?
>>
>> ...
>
> If we are taking it on trust that only human intelligence can solve one of
> these "spot the hydrant" (etc) puzzles, does it not follow that there must be a
> rooom full of humans somewhere setting them?

Or not.

They just collect the stats for each set of pictures, and mark those
that most people click as valid.

And they can also use that information to train their ai to recognize
shapes.

--
Cheers, Carlos.

On Sun, 21 May 2023 13:51:06 +0200, Carlos E.R. wrote:

> On 2023-05-20 08:43, B. R. 'BeAr' Ederson wrote:
>
> ...
>
>> F-Up set to acf.
>> (Seems to me the one with the closest link of the 3 to the topic.)
>
> Please don't. There are always people that don't read the whatever group
> you set follow up to. I don't read alt.comp.freeware, for instance.

Sigh. Group-spanning discussion was never really acceptable in Usenet.
Some Usenet providers didn't even accept messages with many groups as
target and no follow-up set. Nowadays, most (all?) remaining providers
seems to be more lenient.

If the OP intended to draw know-how from several sources, moderate
crossposting of the question was okay. But (s)he should have set up
a single (follow-up) group for the discussion from the beginning. If
people were interested in that discussion, they should have followed
to that target group. - And read the thread in question, temporarily,
if they did not read that group in general. (Most Usenet reader programs
have filter functions and the like for this.)

With the current question, none of the target groups really fits, though.
And I don't really have an idea of a really fitting one, either. Setting
up a new group would be overkill. (Especially, when following procedures
and not just roughly issuing a "newgroup".)

Usenet going down the dumps for the last couple of years, anyways, we
can leave the discussion spanned over 3 groups, nevertheless. Although
I really have problems to understand, what connection to Google captchas
the group alt.comp.os.windows-10 may have. And alt.comp.software.firefox,
IMHO, only comes into play, when this browser (or one of its extensions)
is the cause of any problems encountered. With alt.comp.freeware there
at least is the connection, that several download sources for freeware
are set up behind captcha walls. So discussing captchas is a way to get
better access to such downloads. - Ultimately, this "connection" is a
stretch as well, though... :-(

BeAr
--
===========================================================================
= What do you mean with: "Perfection is always an illusion"? =
===============================================================--(Oops!)===

In article <acdpjjxqp6.ln2@Telcontar.valinor>, Carlos E.R. wrote...
>
> On 2023-05-21 13:05, Philip Herlihy wrote:
> > In article <u49h5r$37kkg$1@novabbs.org>, vader wrote...
> >>
> >> What's your Google captcha algorithm when confronted with odd shapes?
> >>
> >> ...
> >
> > If we are taking it on trust that only human intelligence can solve one of
> > these "spot the hydrant" (etc) puzzles, does it not follow that there must be a
> > rooom full of humans somewhere setting them?
>
> Or not.
>
> They just collect the stats for each set of pictures, and mark those
> that most people click as valid.
>
> And they can also use that information to train their ai to recognize
> shapes.

So how do they introduce a new image set? (Perhaps that's why you sometimes
have to respond to two sets, even though you were pretty confident the first
time?)

--

Phil, London

On 2023-05-21 17:46, Philip Herlihy wrote:
> In article <acdpjjxqp6.ln2@Telcontar.valinor>, Carlos E.R. wrote...
>> On 2023-05-21 13:05, Philip Herlihy wrote:
>>> In article <u49h5r$37kkg$1@novabbs.org>, vader wrote...
>>>>
>>>> What's your Google captcha algorithm when confronted with odd shapes?
>>>>
>>>> ...
>>>
>>> If we are taking it on trust that only human intelligence can solve one of
>>> these "spot the hydrant" (etc) puzzles, does it not follow that there must be a
>>> rooom full of humans somewhere setting them?
>>
>> Or not.
>>
>> They just collect the stats for each set of pictures, and mark those
>> that most people click as valid.
>>
>> And they can also use that information to train their ai to recognize
>> shapes.
>
> So how do they introduce a new image set? (Perhaps that's why you sometimes
> have to respond to two sets, even though you were pretty confident the first
> time?)

Right, they can just put it to people as the first set.

--
Cheers, Carlos.

On Sun, 21 May 2023 19:54:07 +0200, Carlos E.R. wrote:

>>> And they can also use that information to train their ai to recognize
>>> shapes.
>>
>> So how do they introduce a new image set? (Perhaps that's why you sometimes
>> have to respond to two sets, even though you were pretty confident the first
>> time?)
>
> Right, they can just put it to people as the first set.

When these captcha images first came out, weren't they all zoomed into
street signs and the like where I always thought Google was trying to
refine the metadata that goes with each frame of their street car video.