search

zooniverse / Radio-Galaxy-Zoo

4 stars 5 forks source link

LR's comments and found bugs in FF/Scientific Linux #79

Closed vrooje closed 10 years ago

vrooje commented 10 years ago

I'm thinking it's possible SciLinux is not an urgent pre-launch issue...

Gives me image to classify, click radio contours, done, click IR, done, then shows me image with no markings on it. so i click done again, and this time it gives me the screen that we're supposed to see : radio contours + IR circle, and then I click next, and it takes me back to the same source. so two problems:

same thing happens with next source. image with no markings, have to click done an extra time, then it takes me back to the same source again.

the next source works correctly.

edpaget commented 10 years ago

Do you know what version of Firefox he's using?

edpaget commented 10 years ago

Actually this sounds like your issue #74, which is fixed. I'd be a little surprised if there were Firefox/Linux-specific issues, since that I what I mostly developed it against.

vrooje commented 10 years ago

Yes, no word yet on whether his second check has resolved these issues. Entirely possible he tested this an hour ago or so.

On Tue, Dec 17, 2013 at 4:11 PM, Edward Paget notifications@github.comwrote:

Actually this sounds like your issue #74https://github.com/zooniverse/Radio-Galaxy-Zoo/issues/74, which is fixed. I'd be a little surprised if there were Firefox/Linux-specific issues, since that I what I mostly developed it against.

— Reply to this email directly or view it on GitHubhttps://github.com/zooniverse/Radio-Galaxy-Zoo/issues/79#issuecomment-30764189 .

kinnehora commented 10 years ago

This is Larry Rudnick, DocR on RGZ, kinnehora here on GitHub. The above comments were generated from Firefox 24.0 Our system has recently upgraded to 26.0, and I am now commencing tests with the new version and will update here.

vrooje commented 10 years ago

great! Glad you're on GitHub as well. :)

On Tue, Dec 17, 2013 at 4:22 PM, kinnehora notifications@github.com wrote:

This is Larry Rudnick, DocR on RGZ, kinnehora here on GitHub. The above comments were generated from Firefox 24.0 Our system has recently upgraded to 26.0, and I am now commencing tests with the new version and will update here.

— Reply to this email directly or view it on GitHubhttps://github.com/zooniverse/Radio-Galaxy-Zoo/issues/79#issuecomment-30765276 .

kinnehora commented 10 years ago

With FF 26.0, been through about 10 sources. None of the above problems. However, there is a problem beyond the technical that the infrared images are usually so burned out so as not to be useful -- everything is blended together and you can't see individual IR sources.

kinnehora commented 10 years ago

Spoke too soon. Image ARG0000fhj has a strong radio source at the center. No contours visible and wouldn't allow me to click there (in case the contours were hidden by the source)

kinnehora commented 10 years ago

Image ARG0000erc . This may be user error, but i was unable to "mark another" after doing the first source.

vrooje commented 10 years ago

Re: IR scaling: I noticed this too, but as we're launching in less than 90 minutes it's too late to completely re-do this now. What are our options for post-launch re-scaling of the IR images? How will that affect classifications already collected?

kinnehora commented 10 years ago

Another example like ARG0000fh, where there is a source near the center with no contours. Unfortunately, I never got the screen to "discuss"; it just went on to the next source, so I can't tell you which one it was. In both cases, there was a stronger source in the field, don't know whether that is relevant.

edpaget commented 10 years ago

Thanks, it looks like they have generated contours they're just not loading. Michael and I are trying to resolve it now.

kinnehora commented 10 years ago

Regarding the question from vrooje on IR scaling, we will likely have to throw out the badly blended images. This will be a case by case decision unfortunately. Perhaps someone will come up with something clever, but I would be extremely skeptical (since you guys don't know me, that's my general nature, but still this is potentially deadly for many of the classifications)

vrooje commented 10 years ago

Definitely an issue to fix ASAP, then. It's probably a matter of just finding better scale parameters.

One request: it's best for all individual cases to be put in a collection on Talk (from classification page, click "Discuss", then hover over the image on Talk and click "Collect" to make/add to a collection). But, when doing that, it would be better to not comment on each image that it's badly blended etc.

The reason is just that right now a user's initial impression when they come to Talk is that a member of the science team thinks the project is failing, because of all those comments. Collections don't show up in the same way and are much easier to deal with for science/rejection later anyway.

Thanks.

On Tue, Dec 17, 2013 at 4:48 PM, kinnehora notifications@github.com wrote:

Regarding the question from vrooje on IR scaling, we will likely have to throw out the badly blended images. This will be a case by case decision unfortunately. Perhaps someone will come up with something clever, but I would be extremely skeptical (since you guys don't know me, that's my general nature, but still this is potentially deadly for many of the classifications)

— Reply to this email directly or view it on GitHubhttps://github.com/zooniverse/Radio-Galaxy-Zoo/issues/79#issuecomment-30767769 .

vrooje commented 10 years ago

(PS: likewise with technical bugs. It's much better to collect a list of them and then post them here than to write individual comments on Talk.)

On Tue, Dec 17, 2013 at 4:55 PM, Brooke Simmons vrooje@gmail.com wrote:

Definitely an issue to fix ASAP, then. It's probably a matter of just finding better scale parameters.

One request: it's best for all individual cases to be put in a collection on Talk (from classification page, click "Discuss", then hover over the image on Talk and click "Collect" to make/add to a collection). But, when doing that, it would be better to not comment on each image that it's badly blended etc.

The reason is just that right now a user's initial impression when they come to Talk is that a member of the science team thinks the project is failing, because of all those comments. Collections don't show up in the same way and are much easier to deal with for science/rejection later anyway.

Thanks.

On Tue, Dec 17, 2013 at 4:48 PM, kinnehora notifications@github.comwrote:

Regarding the question from vrooje on IR scaling, we will likely have to throw out the badly blended images. This will be a case by case decision unfortunately. Perhaps someone will come up with something clever, but I would be extremely skeptical (since you guys don't know me, that's my general nature, but still this is potentially deadly for many of the classifications)

— Reply to this email directly or view it on GitHubhttps://github.com/zooniverse/Radio-Galaxy-Zoo/issues/79#issuecomment-30767769 .

kinnehora commented 10 years ago

OK - sounds good - just learning the ropes. No more blended image comments or if there's a special case will use "collection" and I think i've not been "talking" about technical problems, but will be careful.

edpaget commented 10 years ago

@vrooje and @kinnehora and @ttfnrob so I looked through contours of some images think I've figure out what's happening. Julie had told me to remove the bottom level from my contouring script since it would mark a lot of noise. When I add it back in there is a contour around the center of the ARG0000fhj.

vrooje commented 10 years ago

Is the same true of http://radiotalk.galaxyzoo.org/#/subjects/ARG0000g0t ? Have just had that come up without contours but there is clearly a radio source there.

edpaget commented 10 years ago

Does it have to do with the scaling on the Radio images? Are they just showing sources which are that bright at a higher relative brightness than they should be?

vrooje commented 10 years ago

Another test case: http://radiotalk.galaxyzoo.org/#/subjects/ARG000068w (let me know if no further cases are needed)

kinnehora commented 10 years ago

This is a problem with contour generation. Not sure the exact algorithm. Clearly it is working ok for many images, and Julie was correct about removing lowest contours BUT the algorithm likely works differently on radio images from different radio surveys. The problems that I have seen are radio images that were generated from the FIRST survey (I can tell by the appearance of the noise on the image) and it seems like the algorithm isn't going down low enough for FIRST, though it may be fine for others.

ttfnrob commented 10 years ago

Is one solution to simply remove those subjects from FIRST until this is resolved? @parrish @edpaget - is that possible?

On 17 December 2013 17:10, kinnehora notifications@github.com wrote:

This is a problem with contour generation. Not sure the exact algorithm. Clearly it is working ok for many images, and Julie was correct about removing lowest contours BUT the algorithm likely works differently on radio images from different radio surveys. The problems that I have seen are radio images that were generated from the FIRST survey (I can tell by the appearance of the noise on the image) and it seems like the algorithm isn't going down low enough for FIRST, though it may be fine for others.

— Reply to this email directly or view it on GitHubhttps://github.com/zooniverse/Radio-Galaxy-Zoo/issues/79#issuecomment-30769820 .

edpaget commented 10 years ago

Nope it's not about the survey or levels (I spoke a bit soon). There was a size cutt-off where we wouldn't use a contour if it was too small. I didn't scale that with the size of the image, since these are smaller than the ones from the beta.

edpaget commented 10 years ago

We're regenerating the contours now then this should be resolved.