[MEncoder-users] Terrible results with x264

[Grozdan]

On Wed, Dec 29, 2010 at 12:02 AM, belcampo <belcampo at zonnet.nl> wrote:
> James Hastings-Trew wrote:
>>
>> On 10-12-23 11:55 AM, belcampo wrote:
>>>
>>> I would like to learn how dumb I am. My observations:
>>> qp gives the most 'constant' quality, less quality variations than crf
>>> qp encodes 'significantly' faster than crf.
>>
>> You have things a little mixed up. QP does not mean constant quality - it
>> means constant quantizer - its only there for testing purposes and will
>> produce non-optimal output. CRF is what some other encoders (like Handbrake)
>> call "Constant Quality" - it adjusts the quantizers on a frame by frame
>> basis to produce a consistant quality across the video - always what you
>> want.
>
> By using -crf I see this figures come by, it doesn't make the impression to
> me that the quality is 'constant'.
> frame=38285 fps= 35 q=31.0 size=  186156kB time=1531.90 bitrate=
> 995.5kbits/s   frame=38297 fps= 35 q=30.0 size=  186316kB time=1532.48
> bitrate= 996.0kbits/s   frame=38313 fps= 35 q=29.0 size=  186380kB
> time=1532.86 bitrate= 996.1kbits/s   frame=38327 fps= 35 q=28.0 size=
> 186476kB time=1533.44 bitrate= 996.2kbits/s   frame=38342 fps= 35 q=29.0
> size=  186572kB time=1534.02 bitrate= 996.3kbits/s   frame=38359 fps= 35
> q=29.0 size=  186684kB time=1534.78 bitrate= 996.4kbits/s   frame=38376 fps=
> 35 q=27.0 size=  186748kB time=1535.36 bitrate= 996.4kbits/s frame=38391
> fps= 35 q=28.0 size=  186812kB time=1535.94 bitrate= 996.4kbits/s
> frame=38410 fps= 35 q=28.0 size=  186908kB time=1536.70 bitrate=
> 996.4kbits/s   frame=38428 fps= 35 q=28.0 size=  187004kB time=1537.47
> bitrate= 996.4kbits/s   frame=38446 fps= 35 q=26.0 size= 187068kB
> time=1538.05 bitrate= 996.4kbits/s   frame=38466 fps= 35 q=26.0 size=
>  187196kB time=1538.82 bitrate= 996.6kbits/s   frame=38497 fps= 35 q=14.0
> size=  187307kB time=1539.78 bitrate= 996.5kbits/s   frame=38537 fps= 35
> q=16.0 size=  187403kB time=1541.31 bitrate= 996.0kbits/s frame=38558 fps=
> 35 q=31.0 size=  187531kB time=1542.85 bitrate= 995.7kbits/s   frame=38574
> fps= 35 q=27.0 size=  187691kB time=1543.42 bitrate= 996.2kbits/s
> frame=38610 fps= 35 q=13.0 size=  187825kB time=1544.34 bitrate=
> 996.3kbits/s   frame=38648 fps= 35 q=15.0 size= 187921kB time=1546.11
> bitrate= 995.7kbits/s   frame=38663 fps= 35 q=31.0 size=  188145kB
> time=1547.07 bitrate= 996.3kbits/s
>
> I didn't specify a bitrate, but it almost looks like cbr.
> q varies from 31 to 13, which is  a very high variation in my opimion. Am I
> misinterpreting matters ??

This is exactly how crf works. It varies the quantize on a per frame
basis, so it may decrease it if it "thinks" a frame needs a lower
quantizer to look good and increase it if it "thinks" that a higher
quantizer on a specific frame will give the same quality as on a
previous frame, which may have required a lower one.

qp just slaps one quantizer on all frames, regardless if a frame may
require a (much) lower quantizer to look good, something that qp won't
adjust for, and regardless if a specific quantizer is overkill for a
frame so it'll just waste bitrate for nothing.

>>
>> _______________________________________________
>> MEncoder-users mailing list
>> MEncoder-users at mplayerhq.hu
>> https://lists.mplayerhq.hu/mailman/listinfo/mencoder-users
>
> _______________________________________________
> MEncoder-users mailing list
> MEncoder-users at mplayerhq.hu
> https://lists.mplayerhq.hu/mailman/listinfo/mencoder-users
>