some new rd tests (was:Re: [MPlayer-users] mpeg_quant test)

[Tuukka Toivonen]

I made some more tests
http://www.ee.oulu.fi/~tuukkat/mplayer/tests/readme.html

- pp=lb vs. pp=md deinterlacing.
  http://www.ee.oulu.fi/~tuukkat/mplayer/tests/deinttest1/readme.html
  Conclusion: pp=md is better, although both are bad.

- mpeg_quant vs. default quantization with and without trellis:
  http://www.ee.oulu.fi/~tuukkat/mplayer/tests/deinttest1/readme.html
  Conclusion: default quantization with trellis is best. Trellis doesn't
  increase nor decrease quality with mpeg_quant.

- H.264 JM 7.3 encoder vs. MPEG-4 Mencoder/libavcodec encoder:
  http://www.ee.oulu.fi/~tuukkat/mplayer/tests/h264test/readme.html
  Conclusion: for movies H.264 makes 20-26% smaller files with the
  same quality than Mencoder's MPEG-4, but the JM 7.3 encoder is
  300-400 times slower than Mencoder and therefore too slow to be
  useful for any real encoding.

I hope you enjoy the H.264 tests, running it took some 2-3 weeks.

On Fri, 7 Nov 2003, Michael Niedermayer wrote:

>> Result: mpeg_quant decreased hugely PSNR.
>mpeg_quant + trellis quantization is not a very well tested combination and

So now it is more well tested =)

>btw, if u have some video which shows "trails" when encoded, then i would be
>interrested to get that

not yet, sorry...

p.s. sorry for posting both into ffmpeg-devel and mplayer-users lists,
but the h264 test is meant for the former and the deinterlacing test for
the latter but I was too lazy for writing too emails.