24p25p-NINE - GOP9 Hack for friendly co-existance of 24p and 25p

towi

Please ignore attachments

Update is here: http://www.personal-view.com/talks/discussion/comment/45642#Comment_45642

rikyxxx

It looks very interesting.

I'll test it tomorrow, thanks fot this new patch!

driftwood

Looks good @towi. And nice to see another new option for end users.

towi

updated setf

jide

just tried out set f and sent it directly to my deathchart. write error and batterypull. card was transcend class 10 16gb. what i don't get is how "orion", which is a much more demanding setting (at least the way i understand it), can survive that chart, and other milder settings won't..

edit: shooting in hbr mode only.

towi

@jide

you are talking about HBRp25 I assume. Hmh... the patch survives all my death charts as well as highly detailed realworld scenes with or without panning. On SanDisk cards, though. Try to lower the FSH bitrate to 36M (or 32M) ... should be rock solid.

As to "Orion"... it survives everything on my cl.10 30MB/s cards. The issue with Orion (and other Intra patches) is: it looks great as long as the codec is not pushed to its limits. But when you shoot highly detailed scenes the footage looks compressed and may show a lot of mosquito noise. I-frames go just to around 500K in Orion... either way how much detail you'll try to squeeze into it.

edit: also... don't forget to reformat your cards in camera after loading a new patch!

cbrandin

@towi

I looked at your screen captures. It looks like StreamParser may not properly be identifying trick mode on the timers - either that or some modes will not be right with longer captures. I noticed that in some modes the max TS bitrate is lower than the video bitrate - which can be problematic. I have a few questions:

How long are the clips for the various modes (not what is reported, but how long they actually are)?

Which version of StreamParser are these captures from?

The PCR time is considerably longer than the clip time in some modes, which means that the video bitrate probably can't be sustained. Do those modes begin to go into fallback mode after a while (a minute, or so)? The frame sizes drop in size dramatically when this happens, so it is easy to see. Or, you can run the JM-SP decoder (built into StreamParser) and it will tell you if T4 matrices (fallback matrices) are being used.

By the way, there is a new version of StreamParser available that you might want to start using.

Chris

LPowell

@cbrandin I'm using 4x Trick Mode in Flow Motion v1.11 in 24L, FH, and H modes, and have also noticed that Stream Parser is sometimes confused in its TS bitrate calculation in Time Mode. (I can check this by estimating overall bitrate from filesize/duration.) If Stream Parser's detection methods are correct, Trick Mode appears to be an "overdrive" mechanism that is triggered when bitrate exceeds certain internal trigger levels. When the bitrate drops low enough, Stream Parser will display 24L, FH, and H files as dropping down to 2x Trick Mode, or as non-Trick Mode streams.

towi

@cbrandin

Hi,

I've just shot 3 minutes of both 24H and HBRp25. The streams hold well all over the 3 minutes. But this goes for my revised version (36M instead of 38M for FSH ... to be posted later today). However PCR duration (whatever that might be :-/ ... ) is again longer than the actual duration of the clip (in HBR mode). Please see attachment.

The screenshots in my first post show the correct recording time. 30 seconds each to prepare the screenshots for the forum. The screenshots have been made with the predecessor of 2.6 (2.5 I assume). The bitrate was not correct for all the clips... this is why I also attached the info from Media Inspector of the respective files.

As to "trick mode" and "timers" ... I don't know what these are...?!

Any advice in making the patch more robust would be much appreciated!

towi

Update added to first post ("24p25p-NINE-HQMatrix")
setting with HBR spanning added in first post ("GOP12-AQ4-44M32M24M-span")

cbrandin

Thanks for the info. "Trick Mode" is when the codec messes with the TS and PCR timers to make things work. What they do is skew these timers so that the reported time is not the same as the actual time (i.e. the elementary stream time). Sometimes they double or quadruple these timers - I already found those. It appears that at 50i they multiply these timers by 6/5. I'll have to look into this. The upshot is that with the HBR capture you posted, the actual PCR duration is probably actually about 3:04 (which is sensible, being only slightly longer than Clip duration), and the TS max speed is probably actually 46337424.

Can you do me a favor? Capture in FSH and SH modes and see if the PCR and Clip durations agree (i.e. whether PCR duration is only very slightly longer than clip duration). Also see if TS max speed is higher than total bitrate.

As to stabilizing the settings - I think AQ4 is much too aggressive - try AQ1.

Chris

cbrandin

Lee - are you saying that it will drop in and out of trick mode during a single capture?

The challenge with calculating trick mode is that it's not totally reliable with short clips - the overhead when starting up becomes relatively big so it's hard to make a determination by comparing clip duration and PCR duration. With longer clips it's easier because the overhead is a relatively fixed amount of about 3 seconds (unfortunately, it varies with mode).

Are you saying that sometimes Frames Mode gets the trick mode calculation right while the Time Mode doesn't?

towi

@cbrandin

"Can you do me a favor? Capture in FSH and SH modes and see if the PCR and Clip durations agree (i.e. whether PCR duration is only very slightly longer than clip duration). Also see if TS max speed is higher than total bitrate."

I guess for those tests 30 seconds each will do?

cbrandin

Yes, that would be fine.

cbrandin

It occurs to me that if 50i uses a trick mode that the actual video bitrate might come out 1.2x what you entered in PTool. This might explain various stability issues with PAL settings.

towi

okay, here you've got them all - 24H, HBR, FSH, FH, SH, H.

For reference I've also attached the respective reports from Media Inspector.

driftwood

@cbrandin You'll also find that Etc mode and 80% modes will tax/tack on the bitrate an extra 15-20%.

LPowell

@cbrandin What I've observed is that with certain clips, Stream Parser will report mostly correct times and bitrates in Frames Mode, but will scale the graph incorrectly in Time Mode. In those cases, it also reports an incorrect TS Max Speed in Frames Mode. I haven't seen any cases where it would drop in and out of Trick Mode within a single clip. The clips I've been testing have all been at least 7 seconds long.

cbrandin

@towi Thanks very much. It looks like a trick mode is used for 1080i50, but not 720p50. Weird!

@driftwood How do you mean Etc mode boosts bitrate, can you post an example? Also, can you post an example of the 80% too? How about when you go up in percentages - is the bitrate cut? You will be able to tell because the Clip and PCR durations will be quite different. Be sure to test with clips at least 30 seconds long, though.

Looks like I'll have to some tuning to the trick mode calculations.

cbrandin

@Lpowell Seven seconds is pretty short - I can see how that might not be long enough. I'll see if I can come up with something, though.

LPowell

@cbrandin What's the minimum duration you'd recommend? I usually shoot for 8 seconds as that makes it easy to estimate bitrate from a glance at the file size.

Trick Mode is determined by the FH/H Video Bitrate setting which affects both 1080i50 and 720p50 modes. Most Trick Mode settings for those modes will produce 4X in 1080i50 and 2X in 720p50 modes, or 2X in 1080i50 mode only. I had to set it extremely low to produce 4X Trick Mode in both 1080i50 and 720p50 modes, and it required careful fine-tuning.

cbrandin

Actually, it looks like trick mode is more complicated than I thought - I have to do some testing/research.

LPowell

@cbrandin I suspect that it's also possible to trigger Trick Mode in FSH and SH modes, and perhaps in 24H mode as well. If these modes are tweaked to prevent Fallback mode from occuring, excessively high bitrate excursions can cause recording to fail suddenly on otherwise uncorrupted streams. My hunch is that forcing the bitrate significantly above the Top Setting in FSH and SH modes triggers Trick Mode, which doubles the effective TS Max Speed and allows the transport stream bitrate to exceed file buffer limits. Hard to confirm, though, since nothing is written to the file.

zzap64

@towi - Great settings thanks. Pleased someone is focusing on 25p/30p settings rather than just 24p.

towi

@zzap64

thanks!

I've just did a patch comparison at HBRp25 on a static (!) scene at ISO800 in lowlight, tungsten.

Setting: smooth -2|-2|-1|-2.

Patches used are: 24p25p-NINE, Sanity (modified), my GOP12-span setting from above, Flowmotion (1.11) and Orion.

Here's a side by side comparision of the HBR shots assambled in a QT mov; first 4 seconds uncorrected, then pushing the blacks: http://www.file-upload.net/download-4155438/5patches_tungsten_ISO800.zip.html

And here are the original MTS files:

24p25p-NINE: http://www.file-upload.net/download-4155445/24p25p-NINE_HBR.zip.html

Sanity (modified: FSH=36M / Matrix added): http://www.file-upload.net/download-4155448/Sanity_HBR.zip.html

GOP12-44M32M-span: http://www.file-upload.net/download-4155452/GOP12_span_HBR.zip.html

Flowmotion (1.11): http://www.file-upload.net/download-4155454/FloMo_HBR.zip.html

Orion: http://wtrns.fr/bw45HcqJ1X5DQzT

Attached the Streamparser grabs from the scene.

I'v also attatched a new patch to the first post with new settings for 24p: 24H set to 66M (and 24L to 44M) if you prefer more moderate bitrates for 24p.

LPowell

So my takeaway from this motionless, low-light test is that for HD still photography of wool fabric, all five HBR patches look pretty much the same, regardless of bitrate and GOP-length?

Howdy, Stranger!

Categories

Tags in Topic

Top Posters