BD Rebuilder Beta - Bug Reports Only

[Sharc]

Originally Posted by HWK

I am using preset medium with resolution 1920*1080 and getting 15 fps during encode. However choosing preset slow (which is next available option after medium speed goes to 6 fps)

Major difference I noticed motion estimation goes from hex to star and result in lower speed. If I chose uneven multihex I get 13 fps on same source and result in better quality.

I am still tweaking and see how it goes, currently doing two pass encode and will keep posted.

I have been using preset faster which is almost twice as fast as medium; and the quality deterioration was very minor. This impression may depend on the source and PC monitor though.

[omegaman7]

1080p is certainly adequate for me, presently. But I'll likely be in the market during tax time, for a 3D display. Might as well get a 4K display at the same time I'm probably also upgrading my Nvidia GPU. I believe my encodes would benefit from it

[sonate]

Originally Posted by omegaman7

1080p is certainly adequate for me, presently. But I'll likely be in the market during tax time, for a 3D display. Might as well get a 4K display at the same time I'm probably also upgrading my Nvidia GPU. I believe my encodes would benefit from it

I'm sure it will be nice and wish you the most enjoyment pleasure. I just spent 36K on a complete rebuild for my piano and another 20K for stone walls to terrace my house which is on a fairly steep hill. So not much money left at the moment.

When flat screens first emerged, I was skeptical about the longevity of the backlights and LCD array but they have proven to last far longer then the time many people actually keep them. Far more advanced replacements cost significantly less it seems, so you practically have to give the used outdated one away or trash it.

[jdobbs]

Originally Posted by HWK

Are you using default setting for x265 or something different.

I am not sure if it happens with your source or not, if you try to seek forward or backward picture get corrupted until next IDR frame.

I just finished saving private ryan and noticed video need at least ABR 3000+ to achieve good quality on that particular source.

I am gone try two pass next.

What measure are you using to determine quality? I don't think you can use subjective measurements like "good quality". I'd recommend PSNR. I think you'll find that a bitrate of 2000 Kbs with preset medium will give you approximately the same PSNR value as CRF 23 (in x264 or x265). CRF 23 is generally considered to be very good quality.

I haven't seen any seeking issues yet. Using MPC has some delays when I seek, though. MPC-HC works perfectly.

If you want to compare it and see what gets a similar quality level to x264 at a given bitrate, you need to do a PSNR comparison on both of them. When the value is close -- then the bitrates are comparable.

I'm still doing testing, though.

[HWK]

Okay, I will do PSNR test soon, did you get to try 4K source?

[jdobbs]

Originally Posted by HWK

Okay, I will do PSNR test soon, did you get to try 4K source?

No I haven't tried 4k. I did do some testing with Full-SBS 3D though.

The problem I have now is that I can't use DirectshowSource() to read from an HEVC M2TS so I can reencode after an import, even though it's supposed to work with libde265... I get a crash.

[Sharc]

Originally Posted by jdobbs

If you want to compare it and see what gets a similar quality level to x264 at a given bitrate, you need to do a PSNR comparison on both of them. When the value is close -- then the bitrates are comparable.

Do you need to force --aq-mode 0 for PSNR comparisons being valid? If so, doesn't this take any psy enhancements out of the equation?

[jdobbs]

Originally Posted by Sharc

Do you need to force --aq-mode 0 for PSNR comparisons being valid? If so, doesn't this take any psy enhancements out of the equation?

There's something you could find wrong with any metric... but I think PSNR is the best choice. SSIM might be a little better -- but it takes longer and, in my experience, gives pretty much the same results as PSNR.

I wouldn't modify any settings for testing -- you should use the ones you use most often. That's the only true test.

[soneca]

I know, aq mode can not be used with PSNR.
A test with computer animated cartoon.

x265 - preset slow - crf 22 - 1229.71 kb/s > psnr = 47.847
x264 - preset slow - crf 18 - 2759.39 kb/s > psnr = 47.834

In "theory" we have achieved the same quality at half the bitrate.
I did not know that the X265 had already reached this level.

[HWK]

Originally Posted by jdobbs

The problem I have now is that I can't use DirectshowSource() to read from an HEVC M2TS so I can reencode after an import, even though it's supposed to work with libde265... I get a crash.

Did you try Lav filters and see if it works for you, it has decoder and splitter builtin.

[jdobbs]

Originally Posted by HWK

Did you try Lav filters and see if it works for you, it has decoder and splitter builtin.

They're Directshow filters, or AVISYNTH filters? Where do I get them?

[jdobbs]

Originally Posted by soneca

I know, aq mode can not be used with PSNR.
A test with computer animated cartoon.

x265 - preset slow - crf 22 - 1229.71 kb/s > psnr = 47.847
x264 - preset slow - crf 18 - 2759.39 kb/s > psnr = 47.834

In "theory" we have achieved the same quality at half the bitrate.
I did not know that the X265 had already reached this level.

That's odd. I ran a whole battery of tests. In all my tests the same CRF value gave approximately the same PSNR in x264 & x265 -- but with x265 giving a 30% or so smaller file size. I didn't use "--preset slow", however.

With a CRF of 22 I got an average PSNR of 47.58 with x264 and 47.72 with x265. Interesting that using the slow preset somehow seems to have made the CRF results diverge.

[sonate]

Originally Posted by omegaman7

1080p is certainly adequate for me, presently. But I'll likely be in the market during tax time, for a 3D display. Might as well get a 4K display at the same time I'm probably also upgrading my Nvidia GPU. I believe my encodes would benefit from it

Just crossed my mind, I will be building a new computer as mine is very old. BD rebuilder encodes are flawless but they take forever. Not wanting to hyjack the forum, but any suggestions you or anyone else has for a best bang for the buck CPU and main board chipset which functions well with BD reb will be very much appreciated.

[HWK]

Originally Posted by jdobbs

They're Directshow filters, or AVISYNTH filters? Where do I get them?

Jdobbs, you can find more info on LAV at http://forum.doom9.org/showthread.php?t=156191

I am not sure if it serve purpose but worth a look.

[Sharc]

Originally Posted by jdobbs

That's odd. I ran a whole battery of tests. In all my tests the same CRF value gave approximately the same PSNR in x264 & x265 -- but with x265 giving a 30% or so smaller file size. I didn't use "--preset slow", however.

With a CRF of 22 I got an average PSNR of 47.58 with x264 and 47.72 with x265. Interesting that using the slow preset somehow seems to have made the CRF results diverge.

For a "movie type" test clip I got the same PSNR value of 40.06x (with psy and aq disabled) for:-
a1. x264 --crf 27.0 --preset slow => filesize 14969 kB
a2. x264 --crf 26.9 --preset medium => filesize 16306 kB
b1. x265 --crf 26.9 --preset faster => filesize 8174 kB (55% of a1.)
b2. x265 --crf 28.5 --preset medium => filesize 8579 kB (57% of a1.)

Surprisingly, b1. has visually (on PC monitor) the edge to my eyes, but differences are minor. b2. seems to soften most. I included bilinearresize(1280,720) in all the tests for encoding speed reasons, but I think the PSNR comparison should nevertheless be valid (?).

[jdobbs]

Originally Posted by HWK

Jdobbs, you can find more info on LAV at http://forum.doom9.org/showthread.php?t=156191

I am not sure if it serve purpose but worth a look.

I know libde265 is based on LAV. I was just wondering if there was a better option.

[jdobbs]

Originally Posted by Sharc

For a "movie type" test clip I got the same PSNR value of 40.06x (with psy and aq disabled) for:-
a1. x264 --crf 27.0 --preset slow => filesize 14969 kB
a2. x264 --crf 26.9 --preset medium => filesize 16306 kB
b1. x265 --crf 26.9 --preset faster => filesize 8174 kB (55% of a1.)
b2. x265 --crf 28.5 --preset medium => filesize 8579 kB (57% of a1.)

Surprisingly, b1. has visually (on PC monitor) the edge to my eyes, but differences are minor. b2. seems to soften most. I included bilinearresize(1280,720) in all the tests for encoding speed reasons, but I think the PSNR comparison should nevertheless be valid (?).

Cool. Thanks.

[Kurtnoise]

Originally Posted by jdobbs

I know libde265 is based on LAV. I was just wondering if there was a better option.

Not at all...libde265 has been created from scratch and is completely different from lav filters.

[jdobbs]

Originally Posted by Kurtnoise

Not at all...libde265 has been created from scratch and is completely different from lav filters.

Hmm... when I go into the Video Configuration dialog it says "Select which formats LAV Video should decode:" -- so I assumed it was based on LAV. The splitter configuration and the audio configuration both also reference LAV.

[Kurtnoise]

you meant the dshow filter ? ok, you're right, sorry...I thought you are using directly the library.