[MPlayer-users] Question on converting to DivX4...

[RC]

On Sun, 10 Jul 2005 22:24:12 -0400
Rich Felker <dalias at aerifal.cx> wrote:

> no it is not. the original content is 'half height'. 

Yes, yes...  It's half-height, but it's not displayed as such on a TV.
On an interlaced display, it is properly rebuilt into a full-height
image.

> interlaced video works like this:

Blah, blah.  You don't have to explain to me how interlaced video works.

> no. tfields loses no information, pp=lb loses up to half the
> information, 

Okay, I'm listening.  How does pp=lb lose half the information?  It uses
all the lines in both two fields to generate one frame, so all the
information is still there.  It doesn't throw away half the fields and
interpolate, so I fail to see how you can say it throws away half.

> tfields throws nothing away. playing back video encoded with tfields
> to an interlaced display should look identical to watching the
> original interlaced video...

You can say it "should" all you want, but I've tried to use tfields
every different way you can imagine, and it looks terrible.  Not even
close to a real interlaced signal.  I've tried on a TV-out, on a normal
CRT, etc.  

Objects that should look still, seem rather shaky and unstable.
In particular, things like the bars across the lower part of the screen
on cable news channels look absolutely awful with tfields.  The edges
all around it are obviously flickering, noisy, etc.  Watching the same
image live on TV, those kinds of things look _rock solid_, and very
clear.

It seems ironic (to me) that your recomendation to take care of the
noise is to use a strong tn value for hqdn3d, as even slight temporal
deinoising looks terrible to me.  It causes the picture to smear badly
as objects move, and looks worse than the noise I want to get rid of.
When I really needed to use hqdn3d, I would only use 0 or 1 for
temporal. (Now that lavcopt "nr" is available, I'm much happier)

What an interlaced display does _should_ be identical to what tfields=0
does (no translation/interpolation) in theory, but watch the output of
that, and the jumping/ flickering is obvious to anyone.

> if you look at single frames, THEN FUCKING YES THERE IS ALIASING.
> interlaced video is inherently aliased. as long as you watch it
> smoothly at 60fps, your eye will mask most of the aliasing.

Watching it at full speed, I can see all these artifacts I've
described.  Not slowed down, not just looking at a single frame. Why it
looks so much better on a TV, I don't know.  

pp=lb doesn't mask all the aliasing, but it sure helps immensely.

I appreciate your attempts to help (I don't appreciat being called
ignorant, but never mind that) but there's hardly any debate.  You
don't even need very good eyes to see that the output of tfields (even
on a TV) looks nothing like a normal TV signal.