Talk

gtoledo3's picture
Re: Tool: Iterator

Hmmm, I think there is something wrong with the iterator that the comparison chart (while I'm sure it's correct technically), may not be conveying (?).

There is something that was working in Leopard that has been broken in SL, but I'm not exactly which patch is the culprit yet. It's in this queue/random/iterator setup,

It felt like sending a message in a bottle to nowhere to have posted this to the developer list (the attachment). I guess I could have started with bug reporter, but that feels futile ( I will, but I just wait and do a few bug reports at a time).

Really, if anyone can figure out why it doesn't work in SL, it would be a-ok by me. I will give all of the kudos in the world! I have something that looks like the n-body, but way better (and stable), by using this as part of a Leopard setup... but it doesn't work in SL, sooo, I'm at a loss.

It's iterating the sprites, and making multiple images, but there's not time delaying going on.

Sorry to sidetrack (though I do think it isn't strictly OT since it may flesh out the Leopard vs. SL iterator comparison). It's really cool to see you put this up here. Definite +1.

PreviewAttachmentSize
PointCounterPoint.qtz16.68 KB

cwright's picture
Re: Tool: Iterator

this composition seems to work fine for me under SL and 10.5 -- what exactly are you seeing that's "wrong" ?

gtoledo3's picture
Re: Tool: Iterator

Do you actually get "delayed" frames in 10.6? I kind of tried to explain what I was seeing in my response to smokris, below. For instance, in Leopard, I could get "time blur" fx - so if I was still, but moved just my hands or something, my hands would turn into a total blur. (Like an old Kung-Fu movie effect or something).

In SL, no "frames" ever seem to delay, and they all seem to reference the current video output. So, the queue seems to not be holding past frames at all (maybe it wasn't supposed to?). Maybe my setup only worked in Leopard by accident of something, but it did work on the NVIDIA 9xxx, the IntelX3100, AND the Nehalem, in 10.5.

smokris's picture
Re: Tool: Iterator

gtoledo3 wrote:
There is something that was working in Leopard that has been broken in SL, but I'm not exactly which patch is the culprit yet. It's in this queue/random/iterator setup,

George, could you simplify this and try to come up with a minimal test case that works as you expect in Leopard but doesn't work in Snow Leopard?

From my tests, Queue and Random used in Iterators behave identically whether in Leopard or Snow Leopard.

gtoledo3's picture
Re: Tool: Iterator

Yeah, I'll try to simplify it some... I'm almost at a loss.

What I used to see were actual frames doing a kind of "time blur", like "trails" of past frames with changing transparency. What it seems like now, is that the frames don't "delay/re-order" at all.

If I moved in front of the cam in sl, I would see a bunch of "time blur" frames chasing to "meet" my current position. Now, I just see my image, and every frame references the current video output (seemingly).

So, if I hit a "wiggle" I would expect to see that "wiggle" happen many times, delayed. Now, everything looks like it's not "delaying" the frames... it was kind of like a "multi-tap visual delay" (as opposed to a multi-tap audio delay), if that metaphor makes sense to anyone.

I'll definitely try to come up with something simpler, but really, the only stuff I could think to cut out would be that one sprite that isn't in the actual iterator, and some of the image resize stuff (all kind of incidental).

Thanks for the thoughts guys.

cybero's picture
Re: Tool: Iterator

Brilliantly succinct and useful information, smokris & Kineme.

GT, perhaps this is again a 10.6.2 problem for you , that might also be related to something you have installed that isn't installed by other Premier / Select members.

I only have a Student membership at present, run 10.6.1 and regularly scour my Console since updating to 10.6 [.x] for any bits of code kludge / mismatch.

Your test composition doesn't raise any such problems at the Console level at all.

There are differences 'twixt Leo and Snowy, though GT.

It is time delaying in SL.

It is time delaying even more obviously so in Leopard on my G5 PPC 10.5.x .

I do wonder if the same degree of video stagger I achieved on my PPC is down to the relative slowness of that machine as distinct from my iMac 10.6.x ?

shmem related perhaps.

gtoledo3's picture
Re: Tool: Iterator

Nah, not a 10.6.2 thing, it does it on every version of SL (nice thought, but I wouldn't ever post about a bug on something pre-release... I think that's REALLY bad form, as is posting about bugs on non-clean installs).

If I ever post about a bug, I totally beg of you not to post anything about it being related to any 3rd party plugins or a new seed or whatever (because I'm probably already mildly irritated at testing something half a dozen times, booting different OS versions, etc., to totally confirm "buggage").

So plugins, or pre-release seed probs will never be the solution to my problem, since the first thing I do is to eradicate that possibility :) I am always talking about something that's current OS, and clean install. Even when I do testing on plugins, I take all of the other plugins OUT.

Good thought process though; it's a reasonable line of thought, but it's totally square one, and it would be really embarrassing to say that QC had a bug that was caused by a 3rd party plugin, or by an unreleased seed. I don't ever want to have to live that down ("boy that cried wolf", anyone?).

I really doubt it is actually time delaying in SL (and not just showing a bunch of sprites stacked with varying levels of transparency, with no time blur), but if it is on some systems and not others, it really sucks. I'm not saying that it's not possible, but it won't on at least a couple GPU's/ different computers in SL, and it DID on multiple Leopard computers.

cybero's picture
Re: Tool: Iterator

Well, GT, I was just trying to cover the possible bases.

Rest assured, there shall be no more references on this thread to that which the rest of the Mac Developers [& Mac Users] do still await :-)

I had tried out the composition in SLeo with both the inbuilt iSight and my original iSight.

I then proceeded to take a couple of screen grabs, from my PPC and my Intel machines.

See these video capture files, at 16.6MB, too large to upload directly upon Kineme's server, both set concurrently to Point and CounterPoint , with an Iteration count of 10.

Although the screen capture process somewhat deteriorated the quality of the resulting render, I think that they do prove that time lapsing is occurring in both PPC-Leo [1.31 fps] and Intel-SL [3.33 fps] renders.

Perhaps someone else can be kind enough to run you a render of this on an Intel-Leo.

Hope that helps.

Oh and upon closer examination of the Console log, the PPC evidenced nil problems, whilst the Intel-SL evidenced the following once, but not upon the second run that was captured:

16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTCtrl2
16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTImg2.0
16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTImg2.1

I have seen this before with Skype, when it was not totally Leo compatible. However, it has not proven, in this case, to be reliably reproducible.

gtoledo3's picture
Re: Tool: Iterator

cybero wrote:
Well, GT, I was just trying to cover the possible bases.

Rest assured, there shall be no more references on this thread to that which the rest of the Mac Developers [& Mac Users] do still await :-)

I had tried out the composition in SLeo with both the inbuilt iSight and my original iSight.

I then proceeded to take a couple of screen grabs, from my PPC and my Intel machines.

See these video capture files, at 16.6MB, too large to upload directly upon Kineme's server, both set concurrently to Point and CounterPoint , with an Iteration count of 10.

Although the screen capture process somewhat deteriorated the quality of the resulting render, I think that they do prove that time lapsing is occurring in both PPC-Leo [1.31 fps] and Intel-SL [3.33 fps] renders.

Perhaps someone else can be kind enough to run you a render of this on an Intel-Leo.

Hope that helps.

Oh and upon closer examination of the Console log, the PPC evidenced nil problems, whilst the Intel-SL evidenced the following once, but not upon the second run that was captured:

16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTCtrl2
16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTImg2.0
16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTImg2.1

I have seen this before with Skype, when it was not totally Leo compatible. However, it has not proven, in this case, to be reliably reproducible.

Well, thanks a million for the screengrabs... downloading them now :) BTW, internet does NOT convey tone. Imagine a weary Mad Max in the Thunderdome, with dust and soot covering him, all dried out... but still cracking a smile. Irritated, but still good humored :) Certainly not irritated at you suggesting the plugin thing or that it was a version thing, because "you never know". However, in my case, I'm just letting it be generally known that I'm really paranoid and overly cautious about ever calling anything a bug without doing a total "clean" test, especially since I have many really odd plugins.

Hmm...ok. Looking at the screengrab. Weird! I think you're getting the mirror image thing (which wouldn't be changed in SL)... The time blur is REALLY obvious in Leopard, as in replicants of yourself. This looks like what I'm getting in SL as well, but maybe with not good fps that gives an impression of "something" hahahah! It_really_looks way different on my system(s) in Leopard. The screen grab definitely is not what I expect to see. Oh my. Back to using the accumulator and sample and holds I guess. It was nice while it lasted :)

I appreciate you doing that test very, very much. Let me know when you need something tested on a dif system :)

So, I used to (and still do, in 10.5) get VERY obvious visual "echo"/delay, even in offline render... like this:

Also, in offline render, I totally get the "time trails" in Leopard, but not in SL, so I think it's not a performance/fps thing causing the time trails. I mean, I get around 40fps in either OS... it's just that in SL, no "visual echo".

cybero's picture
Re: Tool: Iterator

I think the low fps on SL & Leo is the result of my running on a single Dual Core Intel and a G5 PPC.

cwright's picture
Re: Tool: Iterator

rather than video, can you perhaps set it up so that there are 5-10 static frames -- that way we're able to compare against consistent inputs/outputs, instead of less exact descriptions (similar to audio -- I still have no idea what the difference is between a "wet" and a "dry" mix, but I do know that if you encode with MP3 and a very low bitrate, the output can sound "wet" (or, more frequently, "twinkling") ;)

gtoledo3's picture
Re: Tool: Iterator

Hmm, good idea. Except that it only works with moving images.

Yet, I'm realizing how to make an obvious test from you saying that... I think (?).

cybero's picture
Re: Tool: Iterator

Just picked up on the following after ruminating over the virtually imponderable galatic images from the Hubble telescope - the one that needed extra gigs to see the stars sharp

Quote:

rather than video, can you perhaps set it up so that there are 5-10 static frames -- that way we're able to compare against consistent inputs/outputs, instead of less exact descriptions

can do and have done, rejigged one of my slideshows

&

gtoledo3 wrote:

Hmm, good idea. Except that it only works with moving images.

Yet, I'm realizing how to make an obvious test from you saying that... I think (?).

well regarding movement, I just interpolated various aspects of the published inputs to the Point Counter Point original and spliced all of that with the slideshow image rig.

See attached file, exemplar.zip.

PreviewAttachmentSize
exemplar.zip1.4 MB

gtoledo3's picture
Re: Tool: Iterator

You're going to hate me. :) I totally can't tell now ... the "stutter" from loading images is giving it a totally different problem that confounds figuring out the time blur. (Though I can tell it's not, I doubt it would make it any clearer to most?... Maybe it does, in which case I'll go ahead and shut my big mouth. To me, it's clear from just looking at a normal vid feed.)

It looks really really nice. I like the source imagery and it as a qtz (minus the loading pauses... I would probably multiplex something like that).

I'm thinking I'm going to just pipe a sphere or sprite moving back and forth in a render in image, and record vid and stills of it in Leopard, then SL. Or some ultra simple still images being multiplexed (like a flip book style sequence of something moving back and forth and showing how the image will be in the current and "last" place in Leopard, but not in SL... or maybe alternating colors, to show that a third color is made on the "blur" frames). Thassagonnabemytest.

Maybe I'll just make a small distinct area of the time blur part, and another area with a small sprite of with a "clean" video feed, and take some footage? That would probably be best...

I'm thinking of also doing dual offline renders (of same footage in 10.5, then 10.6), because that shows that it has nothing to do with fps or anything (as in, it's not "not doing" the time blur because of some great fps or something that makes the blur frames happen ultra quick).

leegrosbauer's picture
Re: Tool: Iterator

cybero wrote:
Oh and upon closer examination of the Console log, the PPC evidenced nil problems, whilst the Intel-SL evidenced the following once, but not upon the second run that was captured:

16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTCtrl2
16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTImg2.0
16/10/2009 15:44:45   Quartz Composer[1815]   Opening shmem segment com.allocinit.CTImg2.1

I have seen this before with Skype, when it was not totally Leo compatible. However, it has not proven, in this case, to be reliably reproducible.

Don't know what it means, but that com.allocinit stuff is referencing something regarding CamTwist. You could make an inquiry in that application's forums, if desired. For me, 'Opening shmem segment' messages normally appear when CamTwist is first opened. Typically it would be additional references that might have significance. Regardless, an inquiry in the CamTwist forums for clarification is an option.

Incidentally, and as I suspect you may have discovered, in 10.6 CamTwist is no longer available as a video source in the Settings of the QC Video Input patch. I did inquire if this was likely to be a permanent circumstance and the response was 'probably'. Unfortunate, that. I had found the previous availability of that source to be useful.

cybero's picture
Re: Tool: Iterator

Funnily enough, although I have an installation of CamTwist, I was not running it. I also got a Console log message of the same type at 23:37 when I re-opened PointCounterPoint again. Also , not reliably reproducible.

cybero's picture
Re: Tool: Iterator

So, given that CamTwist would have been an available video source, it could be an echo due to some prior lineage - a ghost of what once was whenever a video input source is invoked. Interesting.

leegrosbauer's picture
Re: Tool: Iterator

It is an interesting occurrence, isn't it. Might be worth an inquiry with the developer, but I'm going to guess that he's in about the same boat as many of us regarding finding rhyme or reason to the current conditions in OS X 10.6 Snow Leopard. But then, I know better than to be offering guesses on his behalf. You would have to ask.

cybero's picture
Re: Tool: Iterator

Quote:

Incidentally, and as I suspect you may have discovered, in 10.6 CamTwist is no longer available as a video source in the Settings of the QC Video Input patch

Actually, what I found this morning was that, whilst running PointCounterPoint, when CamTwist was booted up, CamTwist was available as a video input source on the video input patch.

Slightly more puzzling , however, is that the following message was found in my Console

17/10/2009 11:07:53   CamTwist[1155]   *** QCPlugIn: Bundle at path "/Users/cybero/Library/Graphics/Quartz Composer Plug-Ins/QuartzBuilderProtocol.plugin" is not a valid Quartz Composer plug-in

Ah, upon closer examination of the Console output & a little bit of sensible consideration, it did occur to me that this plugin should not be in the Plugins folder but should be in the Patches folder. It was, in fact, a duplicate, no doubt placed there by myself in error :-).

After rectifying that minor error, I still find that CamTwist is available, when booted, as a video source within the video input patch.

smokris's picture
Re: Tool: Iterator

Please put QuartzBuilderProtocol.plugin in Quartz Composer Patches, not Quartz Composer Plug-Ins.

cybero's picture
Re: Tool: Iterator

Indeed smokris - already done :-), thanks for the succinct advice though .

Can't understand how that plugin has ended up in there, especially as a copy of the plugin in question was correctly situated where it should be, indeed, even though the option to install the template is made available from the QB menu, QB automatically checks this is installed this upon boot up after the first post installation boot up when the plugin is automatically installed and one has the option of installing the QB template to the Developer/Library [neat BTW] .

I also don't understand why, or how come, such an incorrectly placed plugin did not get flagged in the Console, at least, not until the CamTwist application was booted to check upon the viability or otherwise of that application acting as a video input source, available from the standard video input patch .

I didn't even get the we'll run with this copy rather than try to run two at one and the same time, which is what one can get when an application has a loose end, or an end user, me, myself personally, I guess [pbck], has duplicated a resource base in error.

leegrosbauer's picture
Re: Tool: Iterator

hmmmm. The only CamTwist plugin that I'm aware of is for QuickTime. To the best of my knowledge, there shouldn't be any CamTwist plugins appearing in Quartz Composer related graphics folders. On my Mac, CamTwist.component is located at /Library/QuickTime.

That said, your observation about finding CamTwist available as a source in the QC Video Input patch has my attention. I'll try to make some inquiries about this anomaly because you may have happened upon something nice, although I haven't a clue if it's technically reasonable to expect it to be accommodated.