[MPlayer-dev-eng] Software scaler regression

[Matthias Wieser]

Hi,

Am Montag, 3. November 2003 01:01 schrieb Michael Niedermayer:
> > Between Oct. 12 and Oct. 26 changes went in that made
> > - sws 1 produces errors when you scale to n*8+x  (1)
>
> cant reproduce

Did you try scale=412:352 ?

> > - all software scaler produce worse colors than before (2)
>
> ur comparission is flawed, u cant compare the unscaled yuv->rgb
> converter against the scaled one

I did compare the scaled one from Oct. 12 to the scaled one from Oct. 30
The newer mplayer produces slightly more reddish images.

> i suggest that u run the scaler several times in
> a row to check if there is a worse color drift than before

Thanks, but I did that, too.
After one scaling you nearly can't see the color drift.
After four scalings the color drift is worse than before.

Here are the images with the scaler applied four times:
http://home.knuut.de/MWieser_/sws_color/raw_detailliert_8_2_4x031012.png
http://home.knuut.de/MWieser_/sws_color/raw_detailliert_8_2_4x031030.png
http://home.knuut.de/MWieser_/sws_color/raw_detailliert_8_2_4x_diff.png

Mplayer command:
[...]-vf crop=$size:$offset,scale=560:454,scale=500:416,scale=440:344,scale=312:252

I forgot to mention: All difference images have been
lightened with gamma=3

Regards,
  Matthias