MacroBs in high BPS scenes

[Demi9OD]

My most recent project was encoding Gettysburg. At four hours and 15 minutes of mostly high detail scenes, I decided to go for a 3CD encode. I encoded in both RV10 (EHQ 85, sharpest) and Xvid (QPel, hvs-best, Motion Search Precision 6, VHQ 1). I was thoroughly pleased with the level of detail kept in the Xvid encode, but even though it came out around 100mb bigger than RV10, the high detail scenes showed far too many macro blocks. I have included my screenshots to show the strengths and weaknesses of both encodes. Shots have been resized from 640*352 to 1024*568 using bicubic .75. Please keep in mine these screenshots were taken in Virtual dub, so post processing is not enabled in Xvid. I normally use ffdshow full post processing and add some video noise to help with macro blocks for Xvid. Is there anything I can do with my Xvid codec settings to get the first scene any better here? I'd rather use Xvid if I can get the bitrate to go a little higher on such scenes, it encodes allot quicker and can be played back on more PCs, not many people even have RP10 or Real Alt. installed.


In this scene I think RV10 is much easier on the eyes.

Source
http://s2k.gamedaemons.net/demi/ss/source1.jpg

RV10
http://s2k.gamedaemons.net/demi/ss/rv1.jpg

Xvid
http://s2k.gamedaemons.net/demi/ss/xvid1.jpg


And here the lack of details is VERY apparent in RV10.

Source
http://s2k.gamedaemons.net/demi/ss/source2.jpg

RV10
http://s2k.gamedaemons.net/demi/ss/rv2.jpg

Xvid
http://s2k.gamedaemons.net/demi/ss/xvid2.jpg

[ObiKenobi]

Well the RV10 while it does have less macroblocks also has a quite noticeable level of detail lost as well. What you might want to do with your xvid encode is a low threshold medium strength smoothing with something like msmooth which will preserve a level of detail but will get rid of the grain and really fine detail to improve the compressibility.

Also might wanna try a few of the other high bitrate matrices as well.

[Demi9OD]

All the noise reduction filters look too unnatural to me. While I whole-heartedly agree that the RV10 encode lacks fine detail, it is often more pleasurable to watch as a whole on my 32" TV, I can't really notice the missing detail on the TV, but I can surely notice the macro blocks. I don't really want to smooth the image out as much as I want to redistribute my bits so high bitrate/busy scenes don?t look so ugly.

[ObiKenobi]

Originally posted by Demi9OD
All the noise reduction filters look too unnatural to me. While I whole-heartedly agree that the RV10 encode lacks fine detail, it is often more pleasurable to watch as a whole on my 32" TV, I can't really notice the missing detail on the TV, but I can surely notice the macro blocks. I don't really want to smooth the image out as much as I want to redistribute my bits so high bitrate/busy scenes don?t look so ugly.

Well then try some of the high bitrate matrices and see if they help you out at all, as they are quite nice. And if say you can't notice the smoothing that RV10 is doing to save bits why are you worried about it when encoding with xvid?

[Demi9OD]

At least when I messed around with smoothing filters in the past, they all created undesirable solid texture effects surrounded by detail. Seemed like the threshold for smoothing and non-smoothing was too fine of a line and would create moving blocks of smoothed video that really detracted from the encode.

What other Matrix' would you guys suggest?

edit: Forgot to mention RV10 audio never synchs right. Opening the final encode in Vdub and comparing it to the source the video can lag 1, 2, or even 3 frames behind the source video, resulting in de-synchronization up to 125ms which really pisses me off.

[ObiKenobi]

Originally posted by Demi9OD
At least when I messed around with smoothing filters in the past, they all created undesirable solid texture effects surrounded by detail. Seemed like the threshold for smoothing and non-smoothing was too fine of a line and would create moving blocks of smoothed video that really detracted from the encode.

Try Donald Graft's msmooth for avisynth. It's a great filter. I usually use a setting of about threshold: 3-4 strength: 5-7. This leaves a good amount of detail but will smooth out grain and will help the compressibility immensely.

What other Matrix' would you guys suggest?

You can try Bulletproof's High Quality Matrix and Didée's 6 of 9 as they are a couple of good ones I've used.

[kilg0r3]

BTW, I still don't like Qpel - probabey because I Am using denoisers, but I would highly recommend trellisquant and a VHQ setting of at least 2.

If you use B-frames, it might be a good idea to not let them use too high quants, I usually go for a ratio of 1 and an offset of 1. If you are having a lot of time, you might also want to join the experiments with very high b-frame thresholds und max b-frame setting (=> a lot of b-frames like 8 or so in a row). I this case, however you should assign the the b-frames a quant which is equal or higher than that of the one of the p-frames. For details. look for the 'more b-frames thread'

[Demi9OD]

Started up 4 encodes using hvs-good, better, trellis with best (I didn't use it first time), and bulletproof matrixes. Should be done by the time I get home from work at 6 tomorrow (gotta love that 3.5ghz p4). Whichever one works the best I'll try applying the b-frame modifications.

[Teegedeck]

Hello Demi9OD,

so nice to read you again.

I feel pretty much the same as you when it comes to noise-filtering and high bitrate encodes.

My guess for your encode would be that not using VHQ=4 and Trellis were primarily responsible for the problems the screenshots show. Use [edit: Changing the preceding sentence completely turned around the meaning of this one before...]also adaptive quantization. Throw at it all XviD has to offer. Quarterpel is a good idea, and GMC helps quality a good deal. 2 b-frames at 1.5 ratio and 1 offset of course. Didée's 6 of 9 seems the best choise for sharp encodes at high bitrates. But with badly compressible sources high SixOf NiceHVS is maybe even better.