Sims not registering accurate markers

[aprajita]

From Els:

Julianne and I are already classifying and we both have had the same thing happening to us. We mark a sim and then we get a message that we've missed it. See attached screen print. Julianne has had it happening with over 20 sims. Is this a known issue ?

Julianne: The sim issue only seems to be occurring in Chrome (45.0.2454.99 m), I use a Windows 8.1 OS. Seem to be working okay on Firefox (40.0.3) - apart from my classification count always resetting!

Els: I'm actually at work and here we use Firefox 38.3.0. I seem to recall I had the same problem when the site went live again with Chrome (not sure which version) I'll let you know tonight.

[aprajita]

Moving counter reset problem to another issue

[aprajita]

Is this just an alpha channel mask problem in the interface or does it mean that they get a wrong in the database - as this would then affect the agents for these users and therefore the classifications? Do we know if the positions of the clicks are being registered properly in the database - perhaps Chris Davis could take a look at marker locations for some of the images to make sure?

[drphilmarshall]

I see this problem too, also using Chrome. It doesn't seem to be completely universal though - I did get a congratultaory message too, so it could be that if there is an offset between the alpha image and the marker position, it's not a very big one.

I think this will not affect the agents, since we can register any marker as a "LENS" vote (as opposed to a "NOT" vote) with the method=False option in mongodb.digest. But, it's annoying for the volunteers!

@chrissnyder Would you mind please checking the marker position being recorded against the position of the region of interest in the alpha image?

On Thu, Sep 24, 2015 at 5:41 AM, aprajita notifications@github.com wrote:

Is this just an alpha channel mask problem in the interface or does it mean that they get a wrong in the database - as this would then affect the agents for these users and therefore the classifications. Do we know if the positions of the clicks are being registered properly in the database - perhaps Chris Davis could take a look at marker locations for some of the images to make sure?

— Reply to this email directly or view it on GitHub https://github.com/zooniverse/Lens-Zoo/issues/293#issuecomment-142916845 .

[aprajita]

oops sorry, yes, I forgot that we don't include the x,y position in the lens or not classfication thanks.

[aprajita]

i tried again in Chrome and also only get this message sometimes, and sometimes get the congratulatory message. I'm not sure what's the problem, but agree it's rather confusing for the volunteers.

@chrissnyder did you manage to have a look at this?

A couple of examples where I got the missed sim message even though I marked http://talk.spacewarps.org/?_ga=1.47163431.1484186906.1429653766#/subjects/ASW000286q http://talk.spacewarps.org/?_ga=1.81625367.1484186906.1429653766#/subjects/ASW00020xy

@drphilmarshall - I could check with some guidance but i don't seem to be able to download the images

Do you remember if Anu had a tolerance parameter in the distance between the x,y makr and the mask - perhaps this may be set to the VICS82 rather than CFHTLS value? If I recall correctly i think this was build into the mask rather than into checking the marker position.

[chrissnyder]

Will be looking at this over the weekend.

[drphilmarshall]

Thanks @chrissnyder, you're a star. I meant to mention this from #286 just now - there are some hints there I think.

[aprajita]

Thanks @chrissnyder, as this seems to be affecting a lot of people, we really need to fix this asap before people are turned off from the site/project.

@drphilmarshall do you have any ideas where this comes from so that it's easier to diagnose?

[drphilmarshall]

I put some links to other issues - I suspect it's to do with vics82 settings being used for CFHTLS images. The image sizes were different, so the coordinate system that the markers are using could be wrong.

On Sunday, September 27, 2015, aprajita notifications@github.com wrote:

Thanks @chrissnyder https://github.com/chrissnyder, as this seems to be affecting a lot of people, we really need to fix this asap before people are turned off from the site/project.

@drphilmarshall https://github.com/drphilmarshall do you have any ideas where this comes from so that it's easier to diagnose?

— Reply to this email directly or view it on GitHub https://github.com/zooniverse/Lens-Zoo/issues/293#issuecomment-143540416 .

[aprajita]

Hi @chrissnyder I hope you don't mind, but I was looking at https://github.com/zooniverse/Lens-Zoo/blob/a8ca72585a958a1824d8278f0865579b36655b0f/app/controllers/classifier.coffee#L364 which has a 10 pixel hack for the annotation position (due to a missing line in the interface I think) and was wondering if this is part of this sim message issue?

As well as determining the location of the success message, y is also being used to test for the feedback message (see line 432), and thus it might be 10 pixels off from the mask in y -

As there's a tolerance in anu's mask, it could be that some inaccurate hits still stay inside the good regions of the mask but sometimes they are out - this may explain the mixture of success and non-success messages

Do you think this might be the cause of the mismatched marker & sim feedback messages...?

I see the same thing is done in the VICS82 branch so am not sure why we didnt have such problems then (if this is the cause).

[chrissnyder]

Maybe, but there also should be plenty of margin of error in the detection, so 10px in any direction shouldn't matter.

I made some changes in e27f571287fa25dee1e05c8e98614596e4c0d493 which fixes this problem under certain conditions. Not sure if I got everything.

[drphilmarshall]

Great, thanks @chrissnyder! I see correct messages now, at least in the short test run I just did. I posted to the discussion at http://talk.spacewarps.org/?_ga=1.203581264.1585278771.1428907215#/boards/BSW0000007/discussions/DSW0000epy

• could be good to monitor this to see if everyone else's problems are solved too.

Thanks again!

On Tue, Sep 29, 2015 at 11:28 PM, Chris Snyder notifications@github.com wrote:

Maybe, but there also should be plenty of margin of error in the detection, so 10px in any direction shouldn't matter.

I made some changes in e27f571 https://github.com/zooniverse/Lens-Zoo/commit/e27f571287fa25dee1e05c8e98614596e4c0d493 which fixes this problem under certain conditions. Not sure if I got everything.

— Reply to this email directly or view it on GitHub https://github.com/zooniverse/Lens-Zoo/issues/293#issuecomment-144272504 .

[aprajita]

i did a check of >200 images and only got errors on those i missed - thanks for fixing this @chrissnyder!

[aprajita]

closing this as I think it's solved (finger's crossed)...