Can Anything Beat MS-MPEG4 V3 2 CD Encode?

[Dali Lama]

I love the new codecs, Divx 5 and Xvid, don't get me wrong, but I just don't see why they can't beat MS-MPEG4 V3 at High Bitrates (1500+). At such high bitrates of course all three codecs look great, but the sharpness of MS-MPEG4 V3 is unmatched. Doom9 has presented this fact in his codec comparison (using Divx 3.11 aka MS-MPEG4 V3) and I have produced the same results.

The point is, why? I don't understand the technicalities of codecs, but it may be that Microsoft has a very good ME (motion estimation) or it might use non-spec code to achieve this sharpness. If the later is the case, then forget about everything I just said, because MPEG4 needs to be standardized and compliant with the future hardware players.

Trying to start a discussion,

Dali

[dragoman]

Hmmmm....

First of all, the comparison you cite did not include DivX 5, DivX 5 Pro, or the latest build of Xvid.

Sorry to all you Nandub lovers out there, but DivX 3.11 is obselete....it cannot compare to the newest codecs (especially DivX 5 Pro w/ b-frames).

Of course quality is subjective, but the majority of the opinions on this board agree with me....

dragoman

[Dali Lama]

Originally posted by dragoman

First of all, the comparison you cite did not include DivX 5, DivX 5 Pro, or the latest build of Xvid.


Ok, that statement is fair, but I have extensively looked at the "new" codecs Divx 5.01 and Latest Xvid. I am just further backing up my own observations with Doom9's.

Secondly, the new versions of Divx and Xvid do not improve upon sharpness at high bitrates, which is the point I am trying to make.

I feel that Microsoft may have something with their hacked upon and ranked upon codec: MPEG4 V3. True it isn't supposed to be in the .avi format and true it may not be MPEG-4 spec compliant, but I am just talking about its superior sharpness/contrast/clarity under high bitrates.

The question of why is important. I am interested spurring on this discussion, because if Xvid or Divx 5 can achieve better sharpness, which I am sure they can, why not look towards this codec as an example. Of course the source of this codec is not available, maybe some smart people can figure out "how" Microsoft accomplished this.

Keeping the discussion going,

Dali

[-h]

I've found this to be true, when comparing screenshots.

The extra noise added by XviD's MPEG quantization almost bridges the gap, but MSMPEG4V3 is still oddly in front. I'd never use it personally though, the smearing created by it's sub-par motion estimation bugs me greatly.

I'm not sure as to the actual reasons behind its performance, I'll have to look through ffmpeg's source to see how its quantization works compared to "mint" MPEG4.

-h

[Teegedeck]

I thought most of this effect was reached by an integrated low-pass filter? Or was that speculation?

IF there is anything objective to the superiority of ISO-MPEG4 codecs at high bitrates, than it would have to be the argument that it seems you can't stop M$-MPEG4 from quantizing uniformously-coloured areas in a very ugly way. Unlike the too-high quantization on dark scenes that Nandub easily counters.

[Dali Lama]

Originally posted by -h

I'm not sure as to the actual reasons behind its performance, I'll have to look through ffmpeg's source to see how its quantization works compared to "mint" MPEG4.


That's the spirit -h. I hope you find some interesting info. Teegedeck and -h are right about the problems of this codec, but if MS is using a lowpass filter then its definitely benefitting its performance. I know other MPEG-2 Encoders use this.

And now that brings me to another idea. Are any of the features of excellent MPEG-2 Encoders applicable to MPEG-4 Encoding? Features such as ZigZag, Linear..., regulating B and P frames, custom quantization (ie: animation, low bitrate), etc. Has anyone tried to implement Multipass system like CCE?

Great to see the ideas,

Dali

[-h]

That's the spirit -h. I hope you find some interesting info. Teegedeck and -h are right about the problems of this codec, but if MS is using a lowpass filter then its definitely benefitting its performance. I know other MPEG-2 Encoders use this.

The odd thing is, the lowpass filter should be removing this extra detail. It's quite strange.

And now that brings me to another idea. Are any of the features of excellent MPEG-2 Encoders applicable to MPEG-4 Encoding? Features such as ZigZag, Linear..., regulating B and P frames, custom quantization (ie: animation, low bitrate), etc. Has anyone tried to implement Multipass system like CCE?

The zig-zag tweaking is already performed in MPEG-4, in a much better way than MPEG-2 does. Custom quantization matrices is possible with XviD also.

I have no idea how CCE's multipass system works, but I assume it would be quite painful to implement.

-h