How to decide what custom matrix should be used when encoding with HC .021

[jakefree]

Hello, fist would like to thank everyone ahead of time for any help that they may offer. I am not sure most the information I will give here will be necessary but I figure the more information the better.
My source is a divx/xvid file, Virtualdub says decompressor is DX50. the quality is not dvd but is better than vhs and supervcd. Using Avisynth (latest release) with medium denoise using mvtools (mvanalyse+mvdenoise). Encoding with HCenc 0.21 minimum bitrate 5300 and max at 9000 2 pass vbr non interlaced 23.976, will do a pulldown with another app after encode. Okay now to the question.
How do you chose what is the best custom matix to use, is it based on the source video or the type of video that you will make? As of right now I am just doin test with HVSbest matrix with adaptive and the picture qualtity is very good but thought it could be better with alittle more knowledge and tweaking.

[guns1inger]

Honestly, under normal encoding and viewing condition I don't you will see much difference between most of them. I'm sure there will be some that argue differently, but having done some test encodes using the same material at various bitrates, I have found the difference to minimal at best. If you are using very low bitrates then there are benefits to be gained by using one of the low bitrate matrices, but otherwise I'm not sure you will see much change.

Unfortunately I am not at my home machine, so I don't have the notes from the tests I did with me, but they are pretty easy to replicate. Just add a trim statement to your avisynth script, find a nice two minute segment, and do some test encodes using your required bitrate and different profiles. 20 minutes spent on this will give you peace of mind.

[manono]

Of all the matrices included with HCEnc, how did you happen to choose the HVS Best? You just liked the sound of the name, maybe? It's a low bitrate (for DVD) and crummy old XviD matrix. You'd be better off keeping to the default matrix. And for lousy AVI sources, it won't make much difference anyway, which you choose. I could prove that, if you had a good DVD source to reencode, and given a high enough bitrate, you could easily tell the difference between, say, the Fox Home Entertainment matrix and the HVS Best.

[Soopafresh]

"with medium denoise using mvtools (mvanalyse+mvdenoise)"

Test out Mvdegrain2 - example included in the MVtools docs. Much better than Mvdenoise.

[jakefree]

"Of all the matrices included with HCEnc, how did you happen to choose the HVS Best? You just liked the sound of the name, maybe?"

To be honest, yes. When looking at all the possible options for custom matrices I naively made a decision based on the name. Nuenc also had that matrix and it was the only matrix that I could use that would allow the app to open again without having to edit my registry.

Okay so in part you base what matix you use by the bitrate, and what in the long run will be visually appealing.

"Test out Mvdegrain2 - example included in the MVtools docs. Much better than Mvdenoise."

I will do that, I am new to avisynth as well, and have been searching for any websites that people might post scrits from basic to advanced so I could geat an idea of what I am getting into.

Thank you again for the info.

[manono]

Okay so in part you base what matix you use by the bitrate, and what in the long run will be visually appealing.

No, I choose the matrix based entirely on the matrix used by the retail DVD, the average quant of the original, and how much I'll be compressing from the original. I just mentioned bitrate because most people understand that better. In actuality, it's an almost useless figure for determining quality.

I am new to avisynth as well, and have been searching for any websites that people might post scripts from basic to advanced so I could get an idea of what I am getting into.

Basic:
http://neuron2.net/dgmpgdec/QuickStart.html

Script Examples, easy and hard:
http://avisynth.org/mediawiki/Script_examples
http://avisynth.org/mediawiki/Shared_functions
https://forum.videohelp.com/topic99389.html

Read everything in these pages a dozen times or so:
http://avisynth.org/mediawiki/Main_Page

[Soopafresh]

Be aware that Mvtools is a slowwwwww plugin, because it is performing many, many, many calculations. Your Athlon 2400 is going to take a while to process the video with MVtools. The good part is that MVdegrain2 is probably the most natural looking denoiser out there.

avisource("myfile.avi")

backward_vec2 = last.MVAnalyse(isb = true, delta = 2, pel = 2, overlap=4, sharp=1, idx= 1,dct=1)
backward_vec1 = last.MVAnalyse(isb = true, delta = 1, pel = 2, overlap=4, sharp=1, idx = 1,dct=1)
forward_vec1 = last.MVAnalyse(isb = false, delta = 1, pel = 2, overlap=4, sharp=1, idx = 1,dct=1)
forward_vec2 = last.MVAnalyse(isb = false, delta = 2, pel = 2, overlap=4, sharp=1, idx = 1,dct=1)
last.MVDegrain2(backward_vec1,forward_vec1,backwar d_vec2,forward_vec2,thSAD=800,idx=1)

[jakefree]

Wow that is slow, 30 hours to do a first pass on a 108 min video. Seems it is time to upgrade the computer. I did a 1 min test on one of the worst parts of the video and it does look better than MVdenoise.

Thank you Manono for the links, I doubt I will ever be proficient at writing my own scripts after looking at some of the advanced ones.

[Soopafresh]

Jakefree - rarely will you have to actually write a script from scratch. Much easier just to adapt one for your requirements.

Yup, if you want the awesome quality, you have to pay the price. A Duo Core Intel E6600 would be 8X faster than your current processor.

Until then, another great denoiser is FFt3dfilter . It will be much faster than Mvtools, at the cost of some quality and accuracy.

[guns1inger]

I like Convolution3d and PeachSmoother, depending on the occasion But even then, a 2 hour video can take 11 - 12 hours to process..

[ronnylov]

The quantizer matrices are some kind of filter that is used in the encoding process that is used to tell the encoder how much data of different frequencies to keep before the quantization is done. So by switching quantizer matrices you can get different average quantizer at the same average bitrate. If the quantizer gets too high there will be encoding artifacts like blocking or mosquito noise. But if the quantizer matrices are throwing away to much then details of the picture are lost and it may be blurry.

So in general you can choose the matrices to get a balance between loss of details and encoding artifacts. What is best to use depends of how difficult this particular source video is to encode and what your average bitrate is set to. But how should you know?

One way to do it is to do a 1-pass encoding at a constant quantizer value that you know gives good results with the quantizer matrice you have chosen. If the resulting file size is too small then you can change to another matrice that keeps more details of the picture (a matrice that is optimized for higher bitrates) and try again. If the file size is too big then you can change to a matrice that throws away more details (optimized for lower bitrates).

After finding a good matrice then just do a normal 2-pass encode with that matrice and check if the quality is OK.