@LPowell Have you set PRE AF to Q-AF or C-AF?

I was pleasantly surprised about good AFC performance of Leica DG 25mm 1.4. Once stopping down to 2.8 onwards, it's pretty good. I'm gonna try it... tonight.

Initially viewing this week - all variations shot so far are totally lacking in blocking or any (for me) blocking artifacts - pixel peeping (which I find strange - i.e. would you open a dialogue moaning at Johhny Marr's use of the Gibson 335 for Smiths songs, instead of another guitar) etc etc Looking at this hack - loving it, as in nothing is distracting me from what I'm pointing at. Much as we wibble on - in real life broadcast or film, they (as in the people paying you) want the best they can get before the delivery date with zero shit - they really dont care about your bitrate :)

I used a 16gb 95mbs Extreme pro original sand disk and had no problems whatsoever on shorter clips with some panning,movement and lots of details( i just stopped them since i had enough footage for what i needed) i am in pal lands.

edit: i didn't mention that i used only the 720 50p mode and some times with 4000 shutter speed just pushing it to the extreme yet no problems.

@LPowell

at first glance these modified settings seem to fix HBR/25p (needs more testing, of course):

• 1080i50 GOP Table=2, 3, 3, 2, 24, 24

• 1080i50 GOPx2=12

• 1080i Top Setting=92000

@peternap With Flow Motion v2, the only advantage I've seen to using Class 10 SD cards rated higher than 30MB/sec is better reliability in 4GB file-spanning at 100Mbps peak bitrates. (With standard Class 10 30MB/sec cards, file-spanning is supported in FM2's 60Mbps 24L, FH, and H modes.)

The flat-to-busy-scene issue is one of several reliability hazards that high bitrate patches must handle, though it is one of the most severe stability tests. I use very sharp lenses with a variety of SD cards for reliability testing, with ISO 200-6400, midrange apertures, 1/50-1/60 shutter speeds, and exposure levels no more than one-stop overexposed. If you see recording failures with exposure settings outside these ranges, it's very helpful to include those details in your report.

re PAL / HBR25p:

write speed error after a few seconds shooting a static death chart (black&white/high contrast) on both SanDisk 30MB/s (16GB) and SanDisk 95MB/s (32GB). What a pity!

I did some lowlight shots in a big room with just a small lightsource in a corner. The cam shut down once in HBR (Pal) mode after 1.50min at iso 320 with the 20mm pancake at f1.7 (Smooth -2 -2 0 -1). All other HBR shots came out just fine. I did tests from ISO 160 up to 2500, all shots were between 2 & 3 minutes long. The average bit rate was around 55MBit/s for all clips. Then I gave 24H a try (but I need 25fps so bad(!)) at ISO2500 and it looked just great. The bit rate was higher at around 100MBit/s. I was using a SanDisk Extreme 30MB/s Class10 8GB Card. Cheers!

@Karl. Nice report. I would be interested in your 720p60 test results. I am having write errors at bitrates above 55MB.

Meanwhile, I reviewed all 1080p recordings I took from the "moving tree leaves in the sun". Here is what I found:

The results for 24H have an average bit rate of 113MBit/s (without ETC) and 118MBit/s (with ETC), and they look very, very good - visibly better than the GOLGOP3 setting I used before (which was also producing very good results at ~80MBit/s). I spotted only very faint compression artifacts where in the lower midtones of a very fast changing area, horizontal borders between different colored areas became visible for a few frames - not something that would come to the viewers attention at normal playback speed.

The results for VMM 80% have an average bit rate of 105MBit/s, and they, too, look very very good, not visibly different from the 24H results. But there was one glitch in ~15 minutes recorded, where some (ffmpeg based) players report the following error message, and 5 frames show damaged blocks (see attached screenshot):

[h264 @ 0xfd6920]negative number of zero coeffs at 104 36 [h264 @ 0xfd6920]error while decoding MB 104 36 [h264 @ 0xfd6920]concealing 3785 DC, 3785 AC, 3785 MV errors

The strange thing about this glitch is that it resembles a similar glitch I experienced when testing GOP3ZILLA, which also had some damaged blocks in the lower third of the screen - and which also was ignored (and decoded with no visible damage) by other players. Maybe this is not at all setting related (and it is only the second time ever I have seen such a glitch).

Last but not least, I tried the HBR (NTSC) mode, and there I experienced a bad surprise: The average bit rate of the HBR 30p recording was only ~50MBit/s, and thus not too surprising, it exhibited serious compression artifacts - see attached screen shot. The quality of the HBR recording is low for the whole "moving leaves" recording, not just in a few frames of it.

Bottom line for me so far: Thanks a lot for what is probably the best 24fps GOP3 setting today - the quality is really awesome.

I hope the HRB/NTSC quirk can be fixed.

Next I will try to do some 720p60 recordings.

@mozes Thanks for confirming my test results. As you've discovered, the GH2 has a tendency to overreact to rapidly changing exposure and detail levels when you start a recording with a blank scene. The AVCHD encoder appears to use the first P-frame in the initial GOP to predict the bitrate requirements of the subsequent video frames. While that works very well for typical shots with static scenery, it can confuse the encoder when the exposure shifts dramatically a few seconds after the recording starts. Since Flow Motion v2 does not use P-frames in its 1080 modes, I was able to take advantage of this P-frame calibration mechanism in the design of its Scaling Tables, using it to manipulate the encoder's rate control at high bitrates.

Unfortunately, I wasn't able to employ this technique in 720p60 mode, since its 6-frame GOP uses P-frames as well as B-frames. I actually put about a week into an attempt to perfect a GOP-3 720p patch, but its image quality wasn't as consistent as the 6-frame GOP used in FM2. To make it even more challenging, I was able to produce reliable and highly optimized results in PAL 720p50 mode that I'm reluctant to compromise with heavy stability adjustments to 720p60 mode,

@LPowell. Lee, just did a quick test of FM2 with Panasonic 14-140mm lens using Sandisk Extreme 32GB 30mb/s and Transcend SDXC 64GB class 10 cards. 24H,L and HBR worked fine with great quality on both cards. FH and H modes worked fine too. The only problem is with SH and FSH which stops with write errors. I really need the 720p60 SH specially for slow motion. It seems like when bitrate for FSH/SH reaches mid to high 50s then stops with write errors. Bitrates below 50 works fine. Will reducing the top end of the SH/FSH bitrate from 99 to a lower value help with this situation without any noticeable compromise? Thanks for all your hard work After refreshing my browser I noticed your comment on further testing on 720p60 and a possible fix.

@LPowell Yes thats it "shooting brightly lit, high-detail subjects"
Just did some test to reproduce that problem.
If i follow a fast moving object, in my case a dirtbike, and it gos from a "dark" background, to a over exposed one (think on a big jump), and back to the dark background, where the camera does a refocus, there it go's wrong when i stop the recording..

I must say your scaling tables and deblocking ones are just briljant.
i did merge them with my own patch, and now the DC in the Elementry Stream Encoder is not 2 anymore, but 6