x264 after 10 months

[Caroliano]

After almost one year from this (http://forum.doom9.org/showthread.php?t=97075) test, I found Sirber's samples floating by my HD. Then I thought: "How is x264 quality now, after all this time? Can it now has an clear advantage over RV10 for very low bitrate anime?"

Then I decided to make some tests:

The source is the same as in sirber's old test. You can find it here (http://ikaruga.co.uk/~snacky/dts/) together with his old encodes. The only other codec that can be compared there is the RV10 as it don't improved in all this time. The others I dunno how much improved. The bitrate is the same insane 400kbps that is very low for this clip. Xvid can't reach in this even on q31. I made my own settings, as I don't found the settings that sirber used that time:

Max quality: All significant for quality options enabled. Even 3-pass becose IIRC it is useful for action-packed very low bitrate short clips like this.

--pass 3 --bitrate 400 --stats "C:\Test2.stats" --keyint 300 --ref 5 --mixed-refs --no-fast-pskip --bframes 5 --b-pyramid --b-rdo --bime --weightb --direct auto --filter 1,1 --subme 7 --trellis 2 --analyse all --8x8dct --me umh

Normal quality: Setings that I normaly use for encoding.I used CRF here because that is what I realy use and because I wanted to see how this "new feature" behave in that extreme situation. I used max-qulity's quantizer as a base to get an bitrate close to 400kbps.

--crf 35 --ref 4 --mixed-refs --no-fast-pskip --bframes 3 --b-pyramid --b-rdo --bime --weightb --filter 1,1 --subme 6 --analyse all --8x8dct

The test results (http://www.badongo.com/file/653262)

x264 improved greatly since then. Now it is also cleary better than RV10 always. Thanks to all developers of x264 for their great work in this codec!

[IgorC]

The only other codec that can be compared there is the RV10 .....

WHAT? Nero AVC.

[Caroliano]

WHAT? Nero AVC.
The ateme codec surely progressed in meanwhile AFAIK. It would be necessary to re-encode it with the latest test version, that I don't have. Or you are talking about the old nero AVC? It was not used in the past test, but if you want you can make your own encode.

And I forgot to say: I encoded that clip in q45 only for fun and included in the zip. The final bitrate is 138.37 kbps! But it is also almost unwachable....

[foxyshadis]

q45 but the average q seems to be around 48-49 actually. (I've noticed high quant crf tends to use a higher average quant than you ask for. Don't think it happens at low quants.) It is amazing that it looks so good at that bitrate, compared to competing codecs, isn't it?

[Caroliano]

q45 but the average q seems to be around 48-49 actually. (I've noticed high quant crf tends to use a higher average quant than you ask for. Don't think it happens at low quants.)
Indeed. Looking at the log:
x264 [info]: slice I:35 Avg QP:45.51 size: 3733 PSNR Mean Y:28.35 U:38.98 V:39.51 Avg:29.86 Global:28.86
x264 [info]: slice P:1657 Avg QP:48.51 size: 1002 PSNR Mean Y:24.32 U:35.93 V:35.87 Avg:25.90 Global:24.96
x264 [info]: slice B:932 Avg QP:48.89 size: 106 PSNR Mean Y:25.89 U:36.22 V:36.42 Avg:27.42 Global:26.37
Maybe this was because the highmotion that confused the 1-pass rate-control. Now I understand how the bitrate was so low.
It is amazing that it looks so good at that bitrate, compared to competing codecs, isn't it?Yes! And that is why I put it there. But what I said is that I would never encode nothing at that bitrate for serious, as the quality is too low. In the other hand, I could encode at q38~39 (the result of --crf 35 for that clip) if I want an very small size, as the quality was not so bad. The usable quantizer range of x264 is 15~40 IMHO.