[XviD-devel] ME Analysis / dynamic Bframes

[Radek Czyz]

Hi,

gruel wrote:
> why does dynamic bframes setting does _not_ set any bframes when the image
> is (almost) completely static?

Heh, first thing I should say is that b-frame decision is _pure evil_.

Some time ago, people complained (and they were, at least partially,
right) that b-frames make static image blurry. That was mostly true
when the image was static, but there was a person talking... the
person's face was blurry.
So I created a mechanizm that limited b-frames in static scenes.
So people complained that the effect of bframes is too small, they
want smaller filesize, as before.

In the end (which was a short time ago), I kept the static scene
detection, but made it hardly do anything. It is still capable of
taking all bframes from a static scene, but it mostly just limits
them....

After encoding my LOTR dvd, I have to say that I am content with the
decision. The scenes which don't have b-frames are indeed sharper,
and everything seems to work (and look) very well. There is also
b-frame threshold which will give static scenes b-frames back...

I expect to change the decision very soon again, but this is mostly
because I change it all the time and I'm never happy about it. It's
evil.

> Anyway, again my question which branch/head to put it. First one went to
> dev-api-4, nobody complained, but nobody cheered, either.

Indeed, I haven't even compiled dev-api-4 yet... so I didn't cheer.
Also, I don't use h263 quant type - is there something which makes it
more difficult to implement trellis to mpeg quant type?

Regards,
Radek