Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
GH2 video conversion difference
  • Hello,

    I've recently bought a Panasonic GH2 because of the great video quality once hacked (Thank you!). I never really compared the quality to the stock firmware because I hacked it right away, and started testing afterwards.

    I made a extremely difficult testing set-up:

    In a completely darkened room I projected thin white stripes of light on a model, and filmed it with different settings. At first I was a bit disappointed with the results, because allthough the footage looked good, a closer inspection revealed some serious flaws.

    When color grading the grain in the blacks appeared very glitchy, almost like a digital rain... it seemed impossible to remove. But I was also trying out different conversion software, and by accident I noticed it made a huge difference..!

    Pavtube was one of the first programs I tried, and apparently was responsible for the 'digital rain'. because ClipWrap gave a much smoother grain in the blacks..!

    Watch the difference here:

    Http://www.johannez.nl/oud/content/dump/testfootage-web.mov (100MB QuickTime movie, half-size but zoomed in 200% for acurate pixel size)

    GH2 - hacked with Flow Motion, filmed in 24P Cinema mode @ 3200 iso (if i remeber this correctly.. is there some way to read out the meta-data and find exactly which settings were used for a particular shot?)

    unfortunately I found out today about this free plug-in, which eliminates the need for conversion software. Allthough ClipWrap still seems to do a lot better coversion job then Final Cut Pro..! https://eww.pass.panasonic.co.jp/pro-av/support/dload/avccam_impt/agree_e.htm

    Hope this helps..!

  • 22 Replies sorted by
  • @lpowell: I will make samples available to you under secured links after Christmas.

  • @LPowell All files have been re-uploaded. You can now see the various program comparisons. The original MTS file is available for download.

    http://www.personal-view.com/talks/discussion/comment/93108#Comment_93108

  • @LPowell Sorry about that. I have converted to Vimeo Plus membership and am in the process of re-uploading the files. Will let you know when that is done.

  • @Johannez

    So, to sum things up, apparently the "digital rain" can appear in underexposed area's shot by a flowmotion-hacked GH2 and not converted by ClipWrap or After Effects, but by something else... would that be safe to say?

    Yes, but apparently @jpbturbo has now seen similar artifacts appear in Premiere-converted footage using two of Driftwood's patches as well. My suspicion is that this is a latent flaw in certain decoders that is sporatically exposed by some specific combination of H.264 encoding parameters. While the patched GH2 does not stray outside the limits of the H.264 encoding format, it's impractical for a decoder manufacturer to test and verify every possible permutation of those parameters, especially if they don't have samples of high-bitrate test footage available.

  • @LPowell I suspect that part of the difficultly is that the original MTS's have the problem on some people setups and not on others, and finding the interaction that causes that the problem is the tricky bit. You already have an MTS file of mine that has the problem on my setup - if you like, in the new year, I can video my screen during playback in Premiere Pro and that should allow to see pretty definitively what we are talking about?

    I did mean to raise a bug with Adobe about this, since I'm sure the problem is with their software, but I modified my production workflow instead and never quite got around to it.

  • @lpowell

    I'll put some .mts files up on my site later this evening.

  • @fredfred27 Thanks for the link, but it looks like your Vimeo sample footage has all been transcoded into MP4 files. Once that has been done, it's impossible to determine what the original source of the artifacts may have been. In this case, the only type of downloadable footage that is useful for technical analysis are the original MTS files, directly from the camera itself.

  • @LPowell Do you have CS6, and if so, have you seen what happens to the footage? I like your hack, but I cannot import Flowmotion 2.02 files directly in to Premiere or After Effects CS6 without this "digital rain" appearing. Yes, it is a flaw of CS6 and not the hack, but it is definately there. Here is the link to my tests. If people want to see the effect being discussed, they should download these examples from Vimeo and examine the footage from the two CS6 example. The original MTS file is there too.

    http://www.personal-view.com/talks/discussion/comment/93108#Comment_93108

  • Glad to see this is putting some similar experiences into perspective..! So, to sum things up, apparently the "digital rain" can appear in underexposed area's shot by a flowmotion-hacked GH2 and not converted by ClipWrap or After Effects, but by something else... would that be safe to say?

  • @radikalfilm

    I hardly have time to reply to this as I'm on a tight deadline with finishing post. That's why I said it's too late for me to learn about a workflow (as presented by the thread starter) which avoids the "rain" in Flowmotion.

    Up to this point, reports of this post-production decoding flaw have only been able to provide fragmentary and inconclusive evidence. It's been challenging even to unambiguously describe the visual nature of its appearance to those who don't already know what to look for. From your account, it sounds like you've spotted it frequently in your footage. Any standout examples you could make available for download in their original MTS files would be greatly appreciated.

  • @sherwood: I hardly have time to reply to this as I'm on a tight deadline with finishing post. That's why I said it's too late for me to learn about a workflow (as presented by the thread starter) which avoids the "rain" in Flowmotion. It is what it is...the .mov linked above by Johannez fully illustrates what the problem (and solution) is. The lesson, as always - know your tools fully and test your workflow end-to-end before starting out, which I did not.

  • @radikalfilm "The "digital rain" caused by low light exposure nearly ruined 80% of my short shot on Flowmotion 2.02. The rest was shot on Valkyrie, which rocks in any light"

    Interesting observations. Perhaps you could back them up with pertinent details from the production (workflow, settings etc.) and example footage comparing the two hacks.

  • I had in fact transcoded precisely to DNxHD, the "rain" shows up there. I have long removed Flowmotion from my GH2 and I'm too deep in post for alternate workflows. For fairness I would add that the same (apparently decoding via CS6 engine) problem shows up in low light unhacked footage. So it's not that there's something terribly screwy with Flowmotion vs stock, but something is done so much better in Valkyrie (and related) patches. It could be a bug in the Adobe decoder that gets triggered by the stock and Flowmotion matrix but not by other matrix sets.

  • @radikalfilm AME transcodes to DNxHD just fine for me.

    You might also try using the "AnotherGUI" front end for ffmpeg to transcode to prores and see if it can transcode without the rain.

  • The "digital rain" caused by low light exposure nearly ruined 80% of my short shot on Flowmotion 2.02. The rest was shot on Valkyrie, which rocks in any light. The problem is with Flowmotion, not with the Adobe mts decoder. EDIT: I just watched the ClipWrap decode, interesting...there are artifacts (to be expected), but not the rain...I stand corrected. However, the combination of Flowmotion+AME is a problem, and needs at least a warning. Too late for me as I'm too deep in post to revert to the original footage.

  • @lpowell I just checked and some earlier footage that I had filmed using one of Nick's drewnet patches does the same thing in my copy of Premiere Pro CS5.

    Edit, some CM Night footage at 60fps also has some corruption in PPro but not in AE :(

  • @LPowell

    we still have no way to alter the camera's built-in gamma curves, which is the primary factor that determines the perceptual quality of its shadow details.

    Succinctly put, as in your other postulations. When will The GH Series - Definitive Guide be published? :-)

  • @lpowell I definitely get the "digital rain" in premiere pro CS5, I haven't tried AE yet.

    AME can transcode the footage to DNxHD just fine however so it's just how premiere reads the .mts files.

    I'll see if I can post some samples later this week.

  • Ack. I never had a problem in CS5 or 5.5, but I get the "digital rain" in CS6.

  • @jpbturbo Actually, the reports I've read are that it's Premiere Pro CS6 that has a problem, and not CS5 or CS5.5. The traces of the effect that I see in Johannez' Pavtube footage are elusive and occur sporatically on just a single frame at a time. The fact that the effect disappears completely in the ClipWrap footage is very provocative, and strongly suggests a decoding error in the Pavtube footage. The frames where it occurs look fine on their own, and do not contain any blatant digital artifacts. It's only when you compare frame-to-frame macroblock levels.

  • Premiere Pro CS5 has the same problem that Pavtube has with flowmotion footage, the weird diagonal digital rain effect.

    Some of the other hacks don't appear to have the same problem.

  • Thanks for posting your decoding results, the differences between Pavtube and ClipWrap renderings are quite noticeable. To answer your question, there are no metadata tags I'm aware of that would indicate patch settings used in an original MTS file. The best tool for examining these files is cbrandin's StreamParser.

    There are several technical issues to bear in mind in evaluating test shots of this nature:

    1. The GH2 is a consumer camera that has only a limited degree of gamma control. In particular, it lacks a pedestal setting, making it impossible to correct its nonlinear handling of shadow detail.

    2. The GH2's low-level noise ramps up noticeably beyond ISO 1600, and further obscures its shadow details.

    3. The 8-bit encoding of AVCHD macroblocks has a brutal effect on near-black shades, rendering them with exaggerated saturation and much cruder posterization than brighter colors. This where the subjective quality of different decoding, chroma smoothing, and noise reduction algorithms can make a noticeable difference.

    The bottom line is that the post-production quality issue you're exploring is to a large degree a salvage operation. To suit typical consumer applications, the GH2 is designed to crush its noisy near-black details and devote most of its bitrate to encoding bright, well-exposed images. While a good hack can improve the GH2's overall performance, we still have no way to alter the camera's built-in gamma curves, which is the primary factor that determines the perceptual quality of its shadow details.