@euggie2000 The patch parameters in the Flow Motion 100Mbps MJPEG patch were optimized to maintain high bitrates even in low light conditions, which makes it very good at minimizing banding in areas of dimly lit gradients. One of the best tools for evaluating MJPEG image quality is JPEGsnoop:

http://jpegsnoop.en.softonic.com/

The GH2 records a JPG screenshot file along with each MJPEG MOV file. If you drag the JPG into JPEGsnoop, it will reveal the Quantization Tables used to compress the macroblocks in the file. Here's the luminance table the GH2 used to encode the low light sample video screenshot below, which was recorded with an average bitrate of 88Mbps:

02 02 02 02 07 05 11 11

02 02 02 02 05 09 11 11

02 02 05 05 05 09 16 14

02 05 05 07 11 14 20 20

02 05 07 11 14 16 11 23

09 07 11 14 16 20 16 20

05 14 18 16 20 23 20 18

14 16 18 18 23 23 20 20

The most important numbers in this table are in the upper left corner, and are used in quantizing the most prominent details in the image. Smaller numbers produce higher quality details, but require higher bitrates to encode. For comparison, here's the luminance quantization table Photoshop used to encode the same JPG at its Maximum quality setting:

02 02 03 04 05 06 08 11

02 02 02 04 05 07 09 11

03 02 03 05 07 09 11 12

04 04 05 07 09 11 12 12

05 05 07 09 11 12 12 12

06 07 09 11 12 12 12 12

08 09 11 12 12 12 12 12

11 11 12 12 12 12 12 12

These tables are very similar, indicating that the 100Mbps MJPEG patch can produce videos in low light that are comparable to maximum-quality Photoshop JPEG's.

@driftwood frames were 1920x1080, these pictures are the screenshots.. Which Quantum mjpeg settings do you recommend to adjust? I am curious if it's possible to raise DR, is there any way to get some kind of the effect in the Nostalgy mode. RED cameras shots show the better shadow details, is it because of the auto hdr shooting?

Try some of the Quantum mjpeg settings but adjust the frame sizes back to 1920 x 1080 and 720 x 1280. See if you can get further improvements than the 100M setting you used.

Having the SLR-Magic 0,95 Lens for some days, I get problems with banding because of soft backgrounds and gradients. The 1920x1080 MJPEG Patch shows lesser banding compared to to the best AVCHD-Patch. I guess, it is because of higher noise. Almost no blocking/mudding in darker areas with MJPEG!! A BIG MUST: Having the ability to record 16bit PCM in 48.000kHz, or at least AC3. Quicktime player shows 16.000kHz sampling frequency in my MJPEG recordings. So audio is not usable. See: http://www.personal-view.com/talks/discussion/492/eoshd-crusade-on-mjpeg-hacks/p2 All the best, Frame

do you mean that the viera tv stretches it on its own?

I really don't know what you are seeing. look at the picture- straight out of the cam. look at the ball. this is the image the lens produces. there is no wrong aspect ratio in it. it is like this. 2 perfectly unstretched images side by side, but one not as large as the other one, due to mechanic incorrectness of the production process.

Nothing to stretch in post here.

on the 3d lens...
I have it, and its 2 pictures in about 3:4 aspect, which already have the correct aspect ratio, which is horrible and unusable for movies. and to make matters worse: the dividing "line" is not in the center, but some 40 pixels off.. its really sad.

maybe with an anamorphic adapter that is really wide one could at least get 16:9?

@driftwood
Nope, wasn't me. My hunch is the chroma tables are hard wired with the luma tables.

@driftwood
Not sure what you mean, the MJPEG encoder has four patchable sets of Tables and Quantization factors that determine how it encodes each macroblock. Table 4 produces the highest detail and I suspect its elements are all set to one, or something very close to that. Tables 0-3 are dysfunctional and produce blank macroblocks. The first attempts at high bitrate patches on the GH1 used nothing but table 4, and as you might imagine, this type of brute force approach produces very inefficient MJPEG encoding that wastes bandwidth and emphasizes edge artifacts.

@driftwood sorry to be a pain about the quality of the image- but considering how dar it would have been- and the fact that this lens is a Daylight lens --- GOOD JOB!

Forget the piccie quality - its spanmybitchup hi iso, testing out in a v dark cafe - for the record - just trying to check frame sizes.

@LPowell I sort of dont wanna be beaten by this bleedin panny 3D lens until I can at least get a decent workflow going and it seems like your mjpeg stuff could make summit happen. So, lets see if anyone can help.
If you record in AVC 1080p24 you have two split images contained in 1920 x 1080 and squeezed as shown in image. So this is two 960 x 1080 pictures for Left and Right.
The problem is you cant just stick it into a Viera 3D TV cus the resultant image is squashed, its like widescreen acting on widescreen. You cant correct aspect ratio on a 3D image with these TVs.

So I need to rip out each side in an editor OR record in a correct aspect ratio which could mean squeezing even more out of the picture using mjpeg so that will 16:9 or whatever best scales to the TV. What would you suggest? Im wondering if 1440 x 540 might crack it.

@alcomposer yep, there's no simple solution/cheap option, but I'm sure we're going to get a few lens manufacturers attempting summit. Syncing 2 gh2s with identical lenses is not for the faint-hearted.