Flow Motion v1.11 Failsafe Patch with HBR 25p & 50p modes

See the first post of this thread for info and downloadable INI file:

http://www.personal-view.com./talks/discussion/2099/100mbps-flow-motion-v1.11-updated-with-failsafe-pal-2550p-modes/p1

@Angry_C Hmmmm... just found snapshot. Thought it was hidden in the Streaming/Exporting Wizard. But VLC 2.0 seems to have a bug, I have large grey blocks on light areas.

@LPowell. Sorry to set you up, but the described effects are still visible. I did see them before using the transcode. But yes, 5DtoRGB does not seem to be a good workflow for editing. Because blacks are raised and whites are lowered, you´ll get the overall impression of a flat, banding free image. But when it comes to colour correction and reestablishing contrast, banding will -of course- show up again, being even worse to handle. Because the camera hits the sweet spot for banding more precise than the color timer.

@Lpowell : do you remember what SS did you use for the graffiti test? I've got a feeling the footage regains some sharpness when the movement is slower, however I can't stand by that, it's just a feeling.

@LPowell

"On the dark end of the scale, however, I've seen decoded AVCHD files that produce RGB values below 16. Here's an example of a Color Finesse screen in After Effects"

This could as well be a color management or importing issue with AE.

I've shot pure black and pure white and loaded both into Avid (direct MTS import, no transcoding prior to import). Black is at 16 and white at 235 ... as supposed to. In Avid I've never seen blacks below 16 on any shot with my GH2.

oh... and by the way. I suspect the "headroom" in the whites we see occasionally seems to come from sharpening after decoding. When you shoot pure white either with the lens completely defocussed or extremely overeposed you should not see values above 235.

@LPowell

I, too have found similar results to Zaven13 and others - great patch. A question re: your comment "cure for this encoding flaw is a small negative Pedestal adjustment". As you are adjusting all color levels, are you using the luma correction to add negative pedestal in PP, or some other adjustment tool. Great work!

@ehr Sure, check my post at the bottom of page 5:

http://www.personal-view.com/talks/discussion/2099/100mbps-flow-motion-v1.1-update-with-hbr-25p-support/p5

@lpowell, I would question any judgements you are able to make about the GH2's levels by looking at RGB values in After Effects, because you don't know how After Effects is doing its YUV to RGB conversion. My analyses were all on the YUV channels straight from decoded MTS files, with no color conversion. I did my analyses with several different decoders and processing software, to be sure it was right. Every time I saw that black is 16 and white is 235, and the number of pixels below 16 or above 235 was very small and could be explained as compression noise.

@LPowell. As I do more testing, I find the 720p60 SH/H having more frequent write errors after anywhere between 30-50 seconds of recording. The 720p60 bit rates are 1.5 to 2 times higher than 1080i60 and HBR. Additionally, the bit rate difference between SH and H mode is anywhere from 5-10 MB/s. Is there anyway you can adjust the 720p60 settings to slightly lower bit rate for SH and may be bigger difference between SH and H so at least H mode will work all the time? I am using Sandisk Extreme 32Gb 30 mb/s and Transcend 64Gb SDXC class 10 cards. Thanks again for all your hard work.

I did a few tests tonight with this patch, and the results were pretty surprising. I don't have everything with me at the moment, but I'm seeing I-frames of 160,000 bytes at 24H/1080p which is around 1/10th the size they should be. It's a small possibility that applying the patch didn't work correctly and that somehow your settings file didn't take hold, but I'm 95% sure everything was applied correctly. I will reload everything and do another test tomorrow. Smooth, -2,-2,0,-2, 160 ISO.

Attached is the StreamParser 2.5 screenshot:

@csync

I'm seeing I-frames of 160,000 bytes at 24H/1080p which is around 1/10th the size they should be. It's a small possibility that applying the patch didn't work correctly and that somehow your settings file didn't take hold

My results are similar to yours, but I don’t think the I-frames a 1/10th of the intended size. The encoder is adaptive after all. Try shooting a death chart and you’ll see I-frames going up in size (but not 10x, of course).

Lee will correct me if I’m wrong, but I don’t see a problem in your test, csync. The encoder varies the size of I-frames and P-frames, depending on the amount of change between the frames, in order to stay within the set limits.

@csync Yes, the height of the I-frames depends on the subject matter and how it's illuminated. Your first Stream Parser sample shows a close-up of a woman's face, half of whose smooth complexion is cloaked in solid shadow. Unlike Quantum patches, which tend to record a great deal of redundant image data, Flow Motion uses only as much bitrate as the image requires. In 1080p24 mode, highly detailed images can readily boost I-frame height to over 1M byte, close to the maximum I-frame data size possible on the GH2.

In Stream Parser 2.5, you can examine detailed image quality metrics for each frame of a video. In the Tools menu, click Create Video Elementary Stream File, then Decode Elementary Stream File. To speed up the analysis, you'll want to first copy the MTS file to your hard disk. When done, this will open a text window showing a chart of frame compression statistics.

In 1080p24 mode, you can expect QP min-max to range around 20-25, but these numbers only tell half the story (the other half is contained in Flow Motion's custom Scaling Tables). An absolute measure of image quality can be found in the DC column, which shows the base DC quantizer for each frame. These numbers will typically range from 6-8, which is a very fine quantizer granularity, and indicates that the frame is encoded with negligible loss of detail.

@oscillian Can you post some sample footage? Recording time can vary greatly depending on subject matter.