Personal View site logo
Crusade on MJPEG hacks
  • This discussion was created from comments split from: 100Mbps GH2 Low Light MJPEG 1080p Patch.
  • 73 Replies sorted by
  • @LPowell

    AVCHD is far superior to MJPEG in low light.

    This is the case on the GH2 and has been so since the GH13. Here is a demonstration (full screen it). Even at 70Mbit versus 44Mbit, MJPEG is far noisier.



    I don't like to see people waste their time on something that's so obviously as flawed a concept as 'an MJPEG low light patch'.

    If you shoot in low light at high ISOs on the GH2 stick to AVCHD, period.

    Remember also that MJPEG 1080p is basically upscaled 1280x720 and 30p, whilst AVCHD Is true 1080p at 24p. That is a huge difference especially the cinematic frame rate.
  • @EOSHD

    Different users could have different opinions.
    If he likes to use his brand, let him.
    If you'll suddenly will like to do the same no one will say a word :-)
  • Maybe someone will do a comparison between GH2vk MJPEG and AVCHD? Its great to have hi bitrate in MJPEG as an option, but I shoot mostly in 24p, so curious to know how it compares.
  • @EOSHD

    It would be nice if you could do some concrete test to show us all what you are saying. You can't take some conclusion and test from the gh1 and apply it as a definite proof for the gh2. They could and surely are very different cameras. If we follow your logic we should all stop testing the gh2 and use the preset and setting of the gh1 on it.

    For now I don't feel like anyone is loosing their time, there is everything to be discovered in the gh2 and everyone contributing (more so someone with the track record of Lpowell) is valuable to all. Perhaps you are right but we will not know until it has been thoroughly tested.
  • EOSHD: "I am sorry but this is a total brand exercise"

    Thank you, EOSHD, that is a real compliment coming from someone who has demonstrated mastery in the craft.

    While AVCHD is a more sophisticated encoder than MJPEG, many of its most efficient compression techniques rely on coarse quantization of image details that it considers insignificant. Unfortunately, dimly-lit shadow detail is one of the aspects of image quality that Panasonic's AVCHD encoders appear to shortchange. At its heart, the GH2 is a consumer camera tailored to produce pleasing results straight out of the box, rather than a high-precision motion capture device designed for a post-production workflow.

    MJPEG videos tend to have a grainier look due to the relatively finer degree of image filtering used in their quantization tables. Some might even consider MJPEG "noisier" than AVCHD, but that's because of MJPEG's tendency to expose more of the image sensor's raw behavior.

    According to cbrandin's recent investigations, the AVCHD encoder's highest quality quantization factors used for shadow detail are around 18-20. The quantization tables used in the 100Mbps GH2 Low Light MJPEG 1080p Patch apply similar degrees of quantization to properly-exposed areas. However, as illumination decreases, the MJPEG encoder automatically uses finer quality quantization tables in dimly lit frames in order to maintain the high bitrates this patch is optimized for.

    Here are examples of the quantization tables used in two of my MJPEG sample files:

    Monument, GH2 MJPEG 1080p @ 1/60 sec:
    Destination ID=0 (Luminance)
    DQT, Row #0: 7 7 7 7 20 13 33 33
    DQT, Row #1: 7 7 7 7 13 26 33 33
    DQT, Row #2: 7 7 13 13 13 26 46 39
    DQT, Row #3: 7 13 13 20 33 39 59 59
    DQT, Row #4: 7 13 20 33 39 46 33 65
    DQT, Row #5: 26 20 33 39 46 59 46 59
    DQT, Row #6: 13 39 52 46 59 65 59 52
    DQT, Row #7: 39 46 52 52 65 65 59 59
    Approx quality factor = 71.42 (scaling=57.16 variance=202.70)

    Monument, GH2 MJPEG 1080p @ 1/500 sec
    Destination ID=0 (Luminance)
    DQT, Row #0: 3 3 3 3 9 6 15 15
    DQT, Row #1: 3 3 3 3 6 12 15 15
    DQT, Row #2: 3 3 6 6 6 12 21 18
    DQT, Row #3: 3 6 6 9 15 18 27 27
    DQT, Row #4: 3 6 9 15 18 21 15 30
    DQT, Row #5: 12 9 15 18 21 27 21 27
    DQT, Row #6: 6 18 24 21 27 30 27 24
    DQT, Row #7: 18 21 24 24 30 30 27 27
    Approx quality factor = 87.05 (scaling=25.91 variance=44.05)

    The upper quantization table was used to encode the overexposed take of the Monument video, and the lower table was used on the underexposed take. While both tables indicate high quality results, the quantization factors used in the dimly-lit take are less than half as coarse as those used in the brightly-lit take. This is exactly the type of encoder behavior that's desirable for preserving low-light shadow details.

  • @blues better at shooting in upscaled 720/30p with more noise, instead of clean 1080/24p at ISO 1600? Please guys, have a reality check with the MJPEG stuff.
  • I like the "raw data" from the sensor with MJPEG - like film grain from a 35mm Kodak Filmcam
    ;)
  • There are two issues that have been met with cynicism lately - this (MJPEG) and the low GOP efforts. Both camps have valid points even if all you consider is quantization. LPowell has articulated what the issue with AVCHD vs. MJPEG is. There is a similar issue with the low GOP results. I frames have a minimum Q value of 18, and P and B frames have a minimum Q value of 20. Therefore, the higher proportion of I frames you have, the more low Q values are used, and the better low light things will look.

    Happily, I have discovered that most people who post here know what they are seeing and I attempt to test things out methodically before dismissing what people say. It has saved me embarrassment, and I'm sure bad feelings on the part of others as well. The fact is that I have confirmed positive assertions people have made much more often than not.

    Lets give each other the benefit of the doubt - I think we all deserve it.

    Chris
  • Shooting one in 10 mins time with Nokton F0.95 and Canon 50mm F1.2

    Using GOP 6 on the 42Mbit AVCHD mode. Any reason to keep GOP at default?
  • A low GOP will decrease efficiency somewhat, detracting from some other aspect of encoding. Like all things in photography, everything is a tradeoff!

    Chris
  • Testing so far reveals AVCHD still looks nicer in low light than this MJPEG patch. Same camera, same lens, best to choose right settings for the task, no?
  • @brianluce Will do. On my site instead since I don't want this one to go any more off topic or flamey than it already has. I've also first got to finish editing it. I support the hard work on MJPEG research and want to see it continue. Have covered that on the blog before. It's just that there are a lot of users out there under the false impression that it's somehow preferable to use MJPEG for certain projects over AVCHD for better image quality when its strong points are cancelled out by the fact that it noisy in low light, less detailed than full 1080p AVCHD and is none-cinematic 30p. You have to always see the wood through the trees.

    Chris makes a good point about noise reduction software and better preservation of detail at high bitrates. But AVCHD 1080p preserves more detail since it is actually 1080p not 720p upscaled and now at 42Mbit it is even better in low light. So again I don't see the logic in using MJPEG to gain detail in low light or low contrast situations.

    I want to see MJPEG improved, of course I do. Just don't want people to be sucked into the Low Light Patch hype when AVCHD is so clearly better for jobs where low light is a factor.

    This is not meant as an attack on anyone or a AVCHD vs MJPEG debate which isn't really on-topic, it's just a statement of advice.

    If you are going to shoot low light, my advice is rather different from @Lpowells
  • @cbrandin totally agree! But it has to be said. Shooters should choose appropriate settings for the job.

    There's such a huge array of confusing MJPEG patches out there with so many different flavours when 2 or 3 will do and none of them are currently suitable for low light shooting.
  • I'm sorry but I don't see LPowell imposing the MJPEG patch on anyone as if it's gospel. Thank him for giving you an extra option, test it yourself to see if you like it, make a decision about what fits your needs, and if you think it's a BS patch, move on with your fu$*#ing life. Why do you have to be suspicious LPowell is trying to sell you his damn brand.
    @EOSHD You need to slightly back off and assume people are more intelligent. "I don't like to see people waste their time on something that's so obviously as flawed a concept as 'an MJPEG low light patch'" .What if my time doesn't need to be rescued by anyone? Did it ever occur to you that someone might just happen to be interested in different looks ? That someone might not always prefer the cleanest look? That someone might want a slightly grainier, slightly softer image for their project and story? Does it sound absurd that given a free choice, someone would choose s16mm over 70mm if the story calls for it?
    @stonebat - the anamorphic guide was just a mistake, everyone makes mistakes, it's hard not to so we shouldn't be so critical. even though this was an easy one to avoid:"Enable the shooting of widescreen 4K DSLR footage, 3840x1080p"-Lol
    P.S sth tells me that this thread will be cleaned up by VK:)
  • @EOSHD I don't understand why you are committing character suicide? I used to like you and look up to you; Now, for the past few weeks, you've been a total ass with the GH2 hack and LPowell. Is everything ok? Why are you so negative?

    EDIT: I'm not trying to make you feel worse, I'm genuinely worried. I myself have had a "burnout" because of work, I know the feeling. With all good in my heart, maybe you shoud take a few weeks off? You are respected in the community, please don't continue in the self destructive path you've now seem to take.
  • Look. LPowell has a personal grudge. As a result I don't like him very much either. But I don't see what any of it has to do with this topic, the other topic, or even my book!?

    Negative? Read the blog, almost all GH2 hack coverage recently and extremely positive. That's my time and my business I've given. I just think a little less personal attacks and a little more civility could be good.

    I'm entitled to my opinions, like them or not there's no reason to get personal or to fall out over anything.

    If you want to believe MJPEG is helpful for low light, nobody is stopping anyone. I can only offer my advice, which I have done here, and I get flamed people with 'prior' history.

    No tweaking of settings will make MJPEG in it's current form a worthy low light option relative to the results from AVCHD. So why bother? If it is to extend the understanding of the codec then fine, but it doesn't.
  • So it is better in it's current form with this patch than AVCHD 42Mbit at ISO 1600?

    More detail? Less noise?

    I don't think so. Moving MJPEG forward in a significant way will not happen with these kind of patches. I'm sorry but it IS a waste of test time.

    It's like a racing team on the edge of a great leap, with 1 hour remaining at a test session, and deciding instead to fiddle with the wheel nuts.
  • In my opinion, these are not the steps that will lead to improvement. These are the same old bitrate numbers in the E tables tweaked slightly and with a low light tagline.

    If the goal is to up-bitrate to maintain more detail and a finer grain of noise that can be removed afterwards in Neat Video or Magic Bullet De-Noiser then that's cool... But to attempt that in MJPEG 720p mode when AVCHD is true 1080p is rather ill conceived. If it's an overall step forward for fine noise grain and more detail on the GH2 in low light conditions that is the goal here, you won't find it by tweaking the E tables. The breakthrough we need has nothing to do with this.

    If anyone wants to give noise reduction a test in AVCHD mode I have written about the plugins here. They work very well. Noise grain in AVCHD mode is good now we have the 42Mbit patch so I am interested to see how much better the plugins work. http://www.eoshd.com/content/470

    Detail and low noise is the goal. 720p upscaled and a primitive MJPEG codec won't give you this.
  • "He didn't seem THAT confrontational."...as long as you're not one of the blind idiots he's referring to....

    I've just done a quick low light A/B test (artificial flowers indoors, natural window light):

    1600 iso; GH2 AVCHD 35/30 settings (clip was 34mbps) ; MJPEG (clip was 93mbps); 14mm pancake.

    Playing both clips side-by-side (i7 Sandybridge; twin monitors)...my impression:

    MJPEG is excellent...grain about the same for both...detail rendering about the same...color rendition similar (but MJPEG whites seemed slightly cleaner). Overall very usable. I prefer the MJPEG image in this particular low light test.

    This was one test...for my own curiosity. MY conclusion is that MJPEG from the GH2 using these settings is excellent. Please test for yourself and make up your own mind...
  • @brianluce I agree it is incredibly confusing to have all these MJPEG patches out, they're a massive distraction. Filmmakers who ask me on twitter what settings I recommend for the hacked GH1 or GH2 - what do I say? Which LPowell patch? This is what guys at the coal face don't often see. These shooters say they have spent hours reading about MJPEG and totally bought into the tech speak on bitrate when they could simply be shooting optimally in AVCHD at cinematic 24p and 42Mbit from the get go. Met one guy here last year who'd hacked his GH1 and was shooting in MJPEG because all he ever read about the hack was how wonderful MJPEG was. That is the effect of these patches... shooters working on important projects who are not tech-heads (and don't have time to get heavily into testing) shooting with a primitive codec in 720/30p because they've been confused by a flood of irellevant information. Sad really, but true.
  • MJPEG LPowell Low Light test at ISO 3200 with Canon FD 50mm F1.2L.

    - MJPEG has higher noise overall
    - Sadly it also has lower detail so you cannot run Neat Video to win back detail as effectively as you can with AVCHD 1080p
    - MJPEG has pink tint due to less noise reduction in the red channel
    - There's coloured moire on fine detail (Look at the hand grip of 600D and around the EOS badge lettering)
    - It's 30p and I much prefer cinema 24p

    Judging noise from stills is not the whole result, the way the noise fizzes over the image is also relevant. Will post final video results on blog later in the week. For now need to get back to work.

    I want MJPEG improved like you all do, it has fantastic potential but this MJPEG bitrate stuff has been done to death now. This is one helluva distraction guys.

    LPowell: "The MJPEG encoder also excels in difficult low-light conditions where the AVCHD encoder cuts down its bitrate to minimum quality levels."

    In theory but clearly not in reality. Look at the smoothness of the gradated tones over the lens release notch on the 600D in the AVCHD shot compared to how it looks in MJPEG mode with the LPowell Low Light patch.

    AVCHD at 1:1 crop:

    image

    MJPEG at 1:1 crop:

    image

    Full frames attached.
    avchd-iso-3200.png
    1920 x 1080 - 2M
    mjpeg-iso-3200.png
    1920 x 1080 - 3M
  • EOSHD,

    Since we lack true 30p at high bit-rates in either 720 or 1080p, until that is available in AVCHD land this research should be encouraged and it is very useful. I need 30p and the AVCHD interlaced modes suck big time in detail and aliasing.

    I suggest you let this one slide and concentrate your efforts on getting other parts of the hack working. GOP 3 still needs work in 720 60 and we need to get rid of the crap start frames and the last frames also. If lpowel wants to do his thing on MJPEG let him do it, its his time and if he wants to waste all the time in the world on it let him be.
  • Here you are.
  • This is an awesome thread. We're losing sight of the larger scheme here with our pixel-peeping. I know everyone likes to bash Zacuto but in the 2011 Shootout Bob Primes sums it up in the end by saying lenses, cameras, film stocks are all preferences - it's about' what you have to say.' I don't care if it's shot in AVCHD or MJPEG or with a fucking elevator CCTV camera - as long as it works in the end and delivers the message in the story that's all that counts for me.
  • To stay on the topic of preferences...Here's a video that showcases rolling shutter. All filmed on an iphone4. "Shortcomings" can be useful if you think outside of the box. It all boils down to preference. Video just wouldnt work as well or be as pleasing if everything was "perfect".