HCenc 1-pass - sometimes it's undersized

[davexnet]

I'm doing some test encodes, some samples, five to twenty minute clips. For testing I'm typically setting the bitrate to approx 7000 kbps
and the 1-pass result it often up to 10% undersized. But it only happens when using a "high bitrate matrix" such as fox1, or fox2.

One encode I did had a base Q of 5.53 (fox1) and 2.48 (hclow) - they both seem reasonable, but the fox1 encode was 6% undersized
while the hclow encode was accurate.

Does the fact that the encoder is having a hard time with the better quality matrix indicate it can't properly control the bitrate
when constrained by the max setting (9000 in this case) ?

Repeating the encodes using 2-pass yields accurate sizes in either case.

Thanks for any info.

[manolito]

Does the fact that the encoder is having a hard time with the better quality matrix indicate it can't properly control the bitrate
when constrained by the max setting (9000 in this case) ?


Yes, I believe you are right on the money with this assumption. I would be very interested in getting to know how the two encodes (first one using Fox, second one using HCLow) compare visually. I would not be surprised if the FOX encode which is undersized actually looks better than the HCLow encode.

Background:
A low bitrate matrix acts like a low-pass filter. It takes bitrate away from high detail / high motion / very bright parts of the source and gives this bitrate to the more flat and static parts. This is desirable for low bitrates because the human eye detects artifacts at static and dark parts much more easily than some lack of precision in high detail and high motion parts.

Using such a low bitrate matrix at very high bitrates like 7000 kbps does not make much sense IMO. You will lose some detail, but at the same time the saved bitrate will not really improve the static parts, because there already is enough bitrate available.


Cheers
manolito

[davexnet]

I did another 1-pass test with the same matrix but a softer source. 7000 kbps, Base Q 3.8 - this time the encoder hit the bitrate OK.

If a combination of the detailed source, the matrix and the max bitrate is the problem that causes the undersize, the solution is reasonable -
substitute a lower bitrate matrix although (as you say), hclow is going a bit too far.

If you're interested in seeing the comparison, I'll PM you a link

[manolito]

If you're interested in seeing the comparison, I'll PM you a link

Of course I'm interested...