RV9 performance latest Codec shootout

[LiFe]

Just out of interest. I have been a fan of RM for some time before the invention of MPEG-4 and was very interested in the codec shootout.

Never seen anyone put the sildes together like that, very smart work!

There were a few complaints of RM9 dropping details in some scenes, and while I agree that it's performance can be improved I prefer dropped details than encoding artefacts.

IMHO a 'better' codec would *never* put something into the video that wasn't in the origional (ie encoding artefacts, for instance the jpeg-like squares on 16x16 boundaries & Dithering etc) however if bandwidth doesn't facilitate encoding all the information, dropping information (starting with fine details, which add atmosphere to a scene but can't usualy be seen) is a much more acceptable tradeoff.

To put it more succinctly: I prefer a codec to drop information, rather than retain sharpness in more of the image but damage it by introducing artefacts.

Interesting to note that all the codecs still have some ways to go before water, smoke, flying debris, and particularly dark scenes look brilliant.

2 cents.

R
LiFe.

[temporance]

Originally posted by LiFe
To put it more succinctly: I prefer a codec to drop information, rather than retain sharpness in more of the image but damage it by introducing artefacts.My $0,02: Any codec can be made to drop information rather that create artifacts. All you need to do is prefilter the source. E.g. if DivX/xvid creates some blocks at 900kbit/s, chances are it will not create those blocks if you prefilter the source (noise reduction / removal of small details).

The difference with RV9 is that you have no control over soft vs. sharp with a few blocks. You can only have soft.

[Tommy Carrot]

If you use postfiltering in mpeg4 codecs, they will behave similar to RV9 (although RV9 is using in-loop filtering, which is a bit better). But not everyone like the filtered image, for example i prefer not to use it.

[Koepi]

MPEG4 codecs were tested with activated post processing as well, so this doesn't count as argument Tommy.

Regards
Koepi

[Tommy Carrot]

Originally posted by Koepi
MPEG4 codecs were tested with activated post processing as well, so this doesn't count as argument Tommy.

Regards
Koepi

I know this, but as LiFe said:

IMHO a 'better' codec would *never* put something into the video that wasn't in the origional (ie encoding artefacts, for instance the jpeg-like squares on 16x16 boundaries & Dithering etc) however if bandwidth doesn't facilitate encoding all the information, dropping information (starting with fine details, which add atmosphere to a scene but can't usualy be seen) is a much more acceptable tradeoff.

To put it more succinctly: I prefer a codec to drop information, rather than retain sharpness in more of the image but damage it by introducing artefacts.


With postprocessing, the above situation should not occur, so i don't know what's this topic about.

[karl_lillevold]

The topic of blockiness/ringing and detail preservation vs smoother video is a matter of preference. The stills comparison in the latest codec shootout is very well made, but I do think that comparing stills in this way, flicking back and forth, and not shown full-screen, vs moving video full-screen, does influence the comparison in favor of the non-filtered images, since blocks are hard to see, and can appear as a sharpening effect to the images, while for moving video blocks and ringing can be more annoying. Remember, most of these are snapshots from what is normally seen as fast motion. Personally, when I view fast motion video, I tend to get more distracted by moving blocks and noise, than the video being a little smoother than the original, but I am fully aware and understand that others prefer detail preservation even when it means some blockiness.

Since this is such a matter of preference, having an adjustable filter in the decoder could have useful. However, RV9 suffers from the same "problem" as for instance H.264. The filter is in-loop, and more effective than normal post-filters. However, this also makes it non-adjustable on the decode side, since the filtered images are used in the prediction loop. Going forward, I will continue to focus improvements and new developments towards detail preservation. RV9-EHQ with the 2-pass VBR improvements is already greatly improved in this respect since the original RV9..

[Sirber]

cool :D

I,d like to see a clip whitout in-loop :D

[Nibor]

Originally posted by karl_lillevold
The topic of blockiness/ringing and detail preservation vs smoother video is a matter of preference. The stills comparison in the latest codec shootout is very well made, but I do think that comparing stills in this way, flicking back and forth, and not shown full-screen, vs moving video full-screen, does influence the comparison in favor of the non-filtered images ...
Yes, it's a matter of preference.
But doom9 did not only do the still image comparison he watched every scene carefully (in full-screen) and that's what his comments are based on :sly: