.MTS files from Panasonic GH2 camera show noise artifacts

[skydragon01]

If I load .MTS media files from my Panasonic GH2 camera onto the timeline of Premiere Pro CS6 (Win7 on PC) on some footage I get a wierd noise/artifact appearing in the footage on playback and also if i render the timeline out.

The artifacts look like a 'rain' of digital noise appearing in certain dark areas of the footage (issue doesn't happen on well lit scenes)

If I use Media Player Classic to play back the .MTS footage outside of PP CS6 the footage plays back 100% ok.

If I transcode the .MTS files into DNxHD format using Adobe Media Encoder the problem persists.

However ....if I transcode the .MTS files into DNxHD format using a 3rd party transcoder (5DtoRGB) the problem is resolved and the DNxHD files work 100% ok in PP CS6

It is as though both PP CS6 and also Adobe Encoder are not interpretting the .MTS file structure ok (?) and introducing artifacts.

It seems that some dev work is needed to resolve the handling of .MTS files by Adobe (?)

Can anyone please assist?

Thanks

[DrDave42]

If you downloaded my files, and they look the same in VLC, then that would be an unusual piece of the puzzle, but no one else has made that observation. It is also unusual that CS5.5 works fine, and CS6 doesn't. I'm not disagreeing with you, everyone has different systems, monitors, etc. My examples were made in a very large room with large areas of shadows, which highlight the "rain" or whatever. Outdoors, you don't see it. I don't want to speak for Ann, but she looked at my clips and echoed back rain in one, not the other. That's what I get and a bunch of people who work in my area. If you get the same rain in both of my clips, then I can't explain that.

I appreciate it that she took the time to look at my clips, and I appreciate everyone's opinion here, of course.

I bought CS6 for Speedgrade and hopefully a better AVCHD experience. I have no sour grapes. I found Speedgrade unusable, but I love the new three-way color corrector in V6. So, great upgrade for me. My AVCHD experience needs a workaround for some clips. I do my pullldown in AE, it works perfect. I do my pulldown in Premiere, not as good. So, no problem. I use AE for pulldown, it looks great.

I would prefer it if everything worked perfect, but this is a great product. I only posted this because I spent hours trying to fix this problem, and I wanted to save some people some grief. If you don't get this unusual digital rain, that is way cool. And if it is the camera that is at fault, or the program, you can just use Premiere 5.5, which also has the advantage that it handles the file spanning.

[Jim_Simon]

If you downloaded my files, and they look the same in VLC

Ann did, as reported in post 20.

she looked at my clips and echoed back rain in one, not the other.

She saw them in the original, but not the converted.  That's not unexpected.  It follows the idea that some software shows the artifacts (PP, VLC, QuickTime and Vegas), but not others (my preferred KM Player does not show it).  That seems very odd to me.  But I don't know enough about how the software works to explain it.

[Guest]

I just tried my clips in CS5 and the fault is not there, just normal low level noise.  So CS5 does handle the files ok and CS6 does not.  What ever the reason CS6 is less capable in this particular aspect than earlier versions.

Understanding where errors are creaping in is mine field these days but I think this is worth investigating by Adobe.  As DrDave says, in a lot of shots you wont see the fault, however it stood out like "dogs bits" on a shoot done last week with a speaker in front of a mid coloured painted wall, ie, pretty much constant area of low saturation colour.  The test I did today and posted as above is of our office wall under exposed on purpose the show the fault and it is really really obvious and really unusable.

So having spent the day investigating and following the paths of other people I have a work around which is reassuring however there is a problem which has been introduced in CS6 and I think it is worth looking into the downgrade in performance in this area.  Hey don't get me wrong,I love Premiere CS6, beats the pants of everything else out there but it has its bugs like everything and the speed attraction of native support of MTS files is lost and I am back to pre rendering.  Can we help Adobe investigate and find if not a solution then a reason??

Cheers

Andy

PS DR Dave I jsut checked your clip encoded via CS6, and I can some of the effect but please guys have a look at the clips I did, they are only a few seconds long but the problem is as clear as night and day, you really can't miss it.

[Jim_Simon]

there is a problem which has been introduced in CS6

Again, given that the issue shows up in several other programs, it is almost certainly with the clips themselves.  Which points to a camera malfunction.  More so that not all GH2 will exhibit the artifacts.

[Guest]

Jim.  I see this fault only in PP CS6, I do not see it anywhere else including in VLC.

[NickDriftwood]

Totally wrong.

Its the programs and the way they handle the mts codec on decode. Hacked GH2 footage goes beyond Profile High Level 4 in bitrate. Adobe and other programmes decoder is unble to read the hi bitrate data correctly - probably decoding it to the confines of the lower elements of profile. Adobe, VLC and the likes  should look into it.

WMP users should dowload K-lite encoder pack which has a better decoder than Adobe's.

[Jim_Simon]

WMP users should dowload K-lite encoder pack which has a better decoder than Adobe's.

That's generally bad advice around here, Nick.  K-Lite has a history of messing up Adobe's software.

[Jim_Simon]

Adobe and other programmes decoder is unble to read the hi bitrate data correctly

Not sure that explains it.  The artifacts have shown up for folks even when using the stock 1.1 firmware.  And they don't show up for every camera.

I still believe the issue is in the clips themselves.  Or at least, a combination of clip and software.

[DrDave42]

There are two files that I posted, one is rendered using Cineform, one is rendered by dropping the AVCHD file on the timeline. The one that Premiere CS 6 rendered has the noise, the one that used the intermediary codec and then was rendered does not have the noise. There is no noise in the original, that is, the original is identical to the Cineform.

Most people who look at the two files will see a difference. If it is important to you, as it seems to be, just take a moment to look at the two files.

[Jim_Simon]

There is no noise in the original

Post it.

[Guest]

Jim.  I have done so above with my MTS clip on Mega storage.

[Jim_Simon]

Jim.  I have done so above with my MTS clip on Mega storage.

Yes, you did.  And I did see the artifacts in VLC.  But DrDave hasn't yet uploaded the original MTS.

[Guest]

Hi Jim

so you are saying you downloaded my original MTS clip, played it in VLC and you saw the rain effet throughout the clip?  I see a short bit at the front then it disappears.  If we are seeing different things wit the same clip and VLC then that confuses things.  What ever I render out of PP CS6 I see the effect in every player.  Pre-render via Prores does not.  I am running CS6 Windows 7 64bit.

Nick thanks for the comments but apart from setting PAL/NTSC and disable record lenth limitation in ptools 1.1 I am running stock bit rate.  The effet happened with stock bit rate and set to average 25mbs.

Cheers

Andy

[Jim_Simon]

I see a short bit at the front then it disappears.

Same here.  But it is there.

[Guest]

Hi Ann and Jim

Yes that is exaclty what I get 1 cycle (25frames) of rain then it is fine.  If I had uploaded more VLC contunies to play without rain.  At least we are seeing the same thing.  I wasn't sure if you always get a disturbance at the start of playing 264 given the nature of the compression.  If I upload more you will see the rain effect building over the 25 frames before it disappears (next key frame?) and then builds again.  It continues cyclically ad ifinitum.

Cheers

Andy

[Jim_Simon]

I hadn't thought of this being somehow connected to keyframes.  I wonder if anyone using an All-I patch has seen the artifacts.

[Guest]

So if I try a longer or shorter GOP in ptools it would demonstrate if the GOP relates to length of rain cycle?

[DrDave42]

I have posted the files, if you want Scientific proof you will need a double blind setup. I posted two files in which most people will see the difference. Some people will not see the difference (although no one yet has said they don't see it).

The reason for the confusion is that the samples are labelled, it is a scientific fact that this skews perception. However, that's the way a forum works. A forum does not arrive at a scientific reality, it attempts to build a consensus (although Lee Smolin argues, and I agree, that consensus building is part of scientific research). The majority of people who look at the clips I posted will see a big, big difference.

So for example you noted that Ann "saw them in the original, but not the converted." My perception is that that is simply not correct, and is in fact backwards. I don't want to speak for anyone, but I mention it because it would seem to show that one clip can be mistaken for another one, even if it is labelled.

But, again, that does not prove anything, If you want a definitive test, you have to set the parameters and use established, scientific methods.

So, yes, I can post my 4 Gig original MTS file, which looks identical to Cineform one I just posted, and then anyone can say, you know, I see some noise in the first frame. That means X. Therefore your camera is the problem. It doesn't matter that in a controlled environment no one could pick out the "winner", what matters, somehow, is getting to some objective reality through non-scientific means.

Then you could of course have another round about a "bug" that creates noise in an intermediary codec--because the noise comes from somewhere, and if it isn't in the Cineform version, then you have simply moved the problem into another sub-forum.

Some people might find that comforting. What I find comforting, is that if I take the same file and render it using a codec, or use CS5.5, the problem goes away. My camera, miraculously, becomes unbroken.

Right now, you can see a consensus building, but you can also see entrenched opinions. That's the nature of a forum, and that's perfectly OK; I always learn something by reading them, and I hope that someone who has this particular problem, whether it is real or imagined, will find a solution that will help them. Some may even be relieved to know that there is at least a possibility that there is nothing wrong with their camera.

I'm assuming by now you have looked at my two clips--and here, I am only talking about my clips, where the problem seems, to me, at least, to be rather obvious. If you decide that the camera likes Cineform but doesn't like Adobe, that's fair, and since without a scientific test it is going to be subjective, it provides a valuable insight for those people who will sift through these comments. If you didn't look at them, well, strangely enough, it results in the same situation.

[Guest]

DrDave.  Yes I looked at your clips, I added mine to hopefully add some extra proof that it is not just your camera and given a plain background the fault really stands out.  With our 3 cameras I think we can exclude a camera fault and I doubt Panasonic's MTS implementation is incorrect, it would be very unlike them.

As you found CS5.5 (or in my case CS5) does not have the problem and CS6 does.  Disregarding any other media players as that is my work flow and the combination of GH2 and PP seems pretty common and getting more common.

It would be great if Adobe jumped in hear and we can all work at what has changed in PP and how we can get a solution that enables direct import of GH2 MTS into PP.  I think the goal is common here, to find the problem and the fix to CS6 to have it process the MTS files the way CS5 did.  It has to be to everybodies advantage surely

Cheers

Andy

[Jim_Simon]

With our 3 cameras I think we can exclude a camera fault

Not if there were a batch of bad models.  It happens sometimes with manufacturing.

if Adobe jumped in hear and we can all work at what has changed in PP and how we can get a solution that enables direct import of GH2 MTS into PP.

Here's the problem, though. Not all GH2's have the artifacts in CS6.  So we need to look at what's different about your clips.  Something clearly is.

[Guest]

Hi Jim

In order to verify if this is a common problem we need the same examples shot on other cameras.  So far we are working on hear say and looking at different types of shots with different content.  There is no way we can say where the issue is without comparing eggs for eggs.

It may be that people just don't see the fault due to the shots they are taking.  I didn't see it for a couple of months with outdoor shots, moving shots etc.

To prove your point are you able to take a short 5 sec MTS clip on your GH2 of a soft coloured surface or wall under exposed by a few stop and then post it for us to download and try?

Surely this would not explain why CS5 plays our clips fine?  There is a diffenrece in the way CS6 and CS5 is handling MTS files.  What is that difference?  There in is the reason for this problem.

[Guest]

Ok, in the interim I have done some further testing.  I set the GOP via ptools to 48 double the standard 24 and found that the cyclic rain jumps and repeats every 48-50 frames, every couple of seconds which is double the time with the standard GOP settings.  So there is a relationship between the I frames and repat of the cycle.

Jim, any chanceo of a plain under exposed MTS from you?

[Guest]

Some further tests...

Tried playing the MTS original clips in some different players.

Windows Media Player 12 - Plays the MTS files fine, no sign even at the front of errors.

BS Player 2.54 - Plays MTS files with rain errors throughtout.

The really interesting thing here is that the BS Player shows a "Main Concept" logo in the playback window so looks like they use the Main Concept filters and as in Premiere this is when the fault happens.  So these test point to the Main Concept filters being the issue.  This possibilty is supported all the more by the fact that if you use FFMB to transcode to Pro-Res, the MTS fiels are read fine and convert without the fault.  I would cetainly guess that FFMBC/FFMPEG does not use Main Concept filters for decoding.

Supporting this is a discussion over on "personal view" which talks about any NLE or player using the current main concept filters having the rain effect.  So if this is the problem does anyone know if we can transfer the CS5 filters over to CS6?

PS Just thought, Dr Dave, do you have the seperate Main Concept transcoder program installed on your system?  I do.  Jim is it installed on yours?  Is there a conflict there maybe?  I will try removing the MC program tomorrow but would be good if you guys can tell me if you have installed on your system.

cheers

Andy

[Jim_Simon]

So these test point to the Main Concept filters being the issue.

Not entirely, as VLC doesn't use them.  Not sure QuickTime does either.  And what about Vegas?

Plus, CS5 does use them.

I have no other MC software installed.

Interesting about the GOP length.  I'll see if I can get some shots tonight.

[Guest]

Does CS5 use a different version of MC filters to CS6?  I don't understand enough to know how PP links to the MC filters.  Is it through direct link to the filters specific to that verson of software or is it system wide generic call to what ever MC filters are installed?

Will still try removing my other MC software today and so if that changes anything.