Official Low GOP topic

driftwood

For those who wanted a default GOP reference, changing the GOP back to 12 (utilising my 100mbps settings from the other night) produces the resulting graph filmed on Pappasdeath chart. AQ=2 seems to be the standard base, AQ3 and AQ4 fall over on the death chart on all my short GOP tests.

plasmasmp

@driftwood

Tried your settings, and I think with AQ set low, there isn't a point to going to 100mb+, you'll just get tiny B frames and a low overall bitrate. I set AQ to 4 and put in my Extreme Pro card and it works great at 140mb. I'm sure it wont span, (haven't tested) but that isn't a huge concern to me.

Nevermind. Looks similar in streamparser. I'll keep testing.

last_SHIFT

@driftwood

Bravo for your low GOP settings. I'm going to try throttle it down to 88 - as I keep getting playback errors when I hit over 100 on my Transcend Class 10 16GB cards. Any settings tips if I lower bitrate?

Cheers

plasmasmp

Ok. For anyone with a Sandisk Extreme Pro, or another card that actually writes 30MB/s plus, try this. What I like: I and B frames nearly exact same size. For maximum stability, I would recommend 3 seconds with lens cap on when starting to record. With that, it will pass a death chart at 132mb AQ4. 126mb 24L should be used for 1080p/ 30 though you will get B-frame issues.

stefanos

just 2 quick framegrabs before&after CC grading from video files created using driftwoods awesome latest settings. iso 3200! F2.8, Smooth -2,0,-2,-1. 1/50 Shutter. WB 3500 K.
Shot with tiny candle in pitch dark, purposefully used highest iso 3200, as I bumped my nikkor 50mm 1.2 to F2.8- could easily have set to 1.4 and shot sth like iso 800:). Graded in color finesse 3 in AE with no noise reduction whatsoever!

I LOVE what I can do with GH2 hacked files! who would have expected I could get such beautiful results and skintones(!!!!!) from iso 3200 with no NR? Does the 1st pic remind anyone of film:) ? Remember all this is a grab from a moving picture with lovely low-GOP motion and quite underexposed. LOOK at the fine details at the skin creases and pores!!!!!!!!
your thoughts?

Big thanks to VK, CBrandin and all awesome testers.

edit: ok I guess this is out of topic here, posted on http://www.personal-view.com/talks/discussion/426/gh2-stalin-samples-topic/p3 instead. VK if u want delete this post.

driftwood

@cbrandin Chris, talk to me about the initial QP setting. This is for the opening frames to establish rate control in the recording, right ?

Edit: I understand this from http://dl.acm.org/citation.cfm?id=1385292: 'To improve video quality and coding efficiency, H.264/AVC adopted an adaptive rate control. But this method has a problem as it cannot predict an accurate quantization parameter (QP) for the first frame. The first QP is decided among four constant values by using encoder input parameters. It does not consider encoding bits, results in significant fluctuation of the image quality and decreases the average quality of the whole coded sequence. '

Therefore can Initital QP setting work alongside the AQ or is it for use alongside manual QP setting only? I'm guessing If unchecked the encoder tries to automatically select the best quantizer for the first picture

towi

@stonebat
"140 is great... but the SDHC card write speed is kinda concern. I didn't have any issue at 130Mbps write speed... but it seems pushing the limit."
Just checked driftwood's latest settings on my SanDisk class 6 cards (20MB/s) ... seem to work without issues

cbrandin

@driftwood

I've played around with the initial QP parameter and it doesn't seem to really make much difference except for maybe the first frame. I have been setting it to the same QP as the other one, but honestly I can't see where it makes any difference at all. The first frame is always small, no matter what I set initial QP to.

Chris

driftwood

@cbrandin thats what Im noticing on Elecard Stream Analyser. I'm currently analysing '100mbps GOP 1' patterns in 'slice_qp_delta ='. Using QP =14 for all tests, Ive set IQP = 0, then 1, then 3 and there is no change in the opening i frame sizes so far. Theyre all hovering around 787500+- in size on streamparser. Unchecking IQP & QP and testing using AQ=2 the size and QP are obviously changing.
e.g. GOP1 test:-

iqp = 0, qp = 14, 1st I frame = 787584 (1st frame = i-10,P-12,i-12,i-12, etc...) testmts000.mts
iqp = 1, qp = 14, 1st I frame = 787776 (i-10,P-12,i-12,i-12, etc...) testmts001.mts
iqp = 3, qp = 14, 1st i frame = 785088 (i-10,P-12,i-12,i-12, etc...) testmts003.mts
Unchecked IQP/QP, use AQ=2 1st i frame = 739968 (i-6, p-13, i-13, i-13, i-13, etc...) testmtsAQ2.mts

cbrandin

@driftwood

The first GOP is always lower quality than the rest. What I mean't was that I found that changing initial QP didn't make a worthwhile difference. At best the QP at the top left macroblock was lower. I've concluded that initial QP isn't really worth messing with because the first frame isn't of much use anyway and past that initial QP isn't used.

Chris

stoney

@cbrandin

Would it be worth to try out 110Mbps, AQ3, and different multipliers on FB1, FB2, and FL?

1.5x failed with AQ3. 1.33x worked... but I didn't check image quality.

It would take only a few more trials. Pick a mid point. Try 1.4x. If worked, try next mid point at 1.45x. If failed, try 1.37x. Repeat a few times to find the highest multiplier that works with AQ3.

But the storm is coming. I can't test it.

driftwood

@cbrandin I think Ive sussed how to get IQP working properly.
This formula works out the IQP
QPy = 26 + pic_init_qp_minus26 + slice_qp_delta.
If I put in IQP = 32, then 26+32-26 then Sliceqpdelta=6 (plus figures at last!)
So my theory is your initialQP has to be over 26 cus everything else was getting minus figures. What you think?
Which means 26 should be the sweetspot bitrate on the first frame youre gonna get. Anything lower in IQP will roughly be the same as 26. Gonna test this.

Edit: putting in IQP=26 gives a slice_qp_delta of 0 in the first frame slice (confirmed by elecard stream analyser).

relevance: Some h264 web sites mention that the slice_qp_delta should be set to 0 in all cases as this value has no bearing on skipped macroblocks.

Mihuel

@stonebat
What you mean "failed". I took the Driftwood settings (140Mbists) changed the bitrate to 110, AQ and PQ = 7 and it works. From here my question.

stoney

The cadence anomaly happened.

Mihuel

@stonebat
Thanks. This is my settings and picture from streamParser.

stoney

If I:B ratio is low like that, something is not right. When taking static scene, 4:1 ratio seems about right as @cbrandin had said.

Mihuel

@stonebat
Thanks for info.I thought the B frame is as big as I frame is B frame retains the characteristics of I frame. Well learn something new. thanks a lot :).It is good that we have these guys like Chris and Vitaliy :)

stoney

+1

Good settings will exhibit static scene 4:1 ratio and moving scene 3:2 ratio that is almost like Intra frames. But that's just ballpark figures. If B-frame begins too high, some weird things happen. I guess it runs out of bitrate bandwidth.

@Mihuel can you test out 110Mbps AQ3 FB/FLx1.4 on highly detailed scene for at least a min?

Mihuel

@stonebat
I'll try this settings today. Give me some time ;)

stoney

Cool. If the rain stops tomorrow, I can test out, too.

Mihuel

@ Stonebat
For me it is already night (and I live on 29m2 with my pregnant girlfriend.) My test consisted of shooting me in the mirror(bathroom) - go after my cat then my sleeping beloved an end of what I see outside the window - a tree, street,tree.

stoney

Hmmm... 1.4x too high. Let me try out 1.37x later.

Mihuel

@stonebat
I like the idea of 3GOP so even try to improve it. If you have any idea it is willing to test it tomorrow :-)

stoney

Once we find the highest multiplier that works with AQ3... we gotta compare image quality with AQ3 and 1.5x.

Ausboy

@stonebat Testing your 140 mbps settings and so far so good. Stable so far - great job - my go to settings now. Thanks!!!!

Howdy, Stranger!

Categories

Tags in Topic

Top Posters