REALmagic MPEG-4 Video Codec

[Chibi Jasmin]

I found this

http://www.sigmadesigns.com/products/MPEG4_video_encoder.htm

page today...a quick search in the forum here made me think, it hasn't been discussed here...I don't have the time to test it, but maybe someone who can test it, would post his/her results? Is it any good? :D

[ulfschack]

Yes, could be very interesting. Come on people ... do the tests so that I don't have to :)

Pros:
Bidirectional frames
Does interlaced

Cons(?):
No multi-Pass


cheers

[Chibi Jasmin]

It features adaptive quantization...if I understand it correctly, it can use different quantizer in different areas of a single frame...

[Latexxx]

I'll test it when I get back home.

PS. Lsx-mpeg encoder plug v2 has also mpeg-4 encoding, I haven't tested it, but maybe I shoul test it too.

[Chibi Jasmin]

Thanx in advance! There seem to be a lot of mpeg-4 codecs popping up...

[Acaila]

It features adaptive quantization...if I understand it correctly, it can use different quantizer in different areas of a single frame...
It means the quantizer matrix can be adjusted for certain macroblocks. Quantizing the frame comes after the quantizer matrix step depending on the size of the frame after the matrix has been applied. XviD's Lumi Masking and DivX5's Psycovisual Modelling are examples of adaptive quantization.

And I'll be checking this codec out when I get home :)

Btw, this codec's settings resemble XviD's quite a bit IMO...

[Chibi Jasmin]

Thanx for details...waiting for results :-) I'll be out the rest of the day...:)

[TactX]

Just did a small test (REALmagic compared to XviD-22052002):

- REALmagic is a bit faster (~9%) :)
- No 2-pass :(
- Final .AVI could not be opened in VDub until I changed the FourCC. :(
- When searching through the AVI in MediaPlayer (also VDub) I got real bad artefacts until the next I-frame ! :(
- Playback with ffdshow was not fluid (not caused by CPU-weakness) :(
- Using Q=4 it looked worse than XviD (Q=3) on "Neo vs. Morpheus fighting-scene" and the file was even bigger ! :( (XviD-decoding by ffdshow but all options unchecked)


what do we learn ?

- 5:1 for XviD => XviD becomes World-Champion :)
- I uninstalled REALmagic :devil:
- Magic doesn't work for MPEG-4, even if it's "real" magic :D

[Latexxx]

I didn't test the video quality yet, but it looks good.
Read the following!
This codec has fastest mpeg-4 decoder that I have ever seen. That makes it also the best from my point of view. I encoded a small clip from M:I-2 (capture via s-video from dvd) using b-frames etc.The resolution was 253*576. When playing it the decoder gives about 25 frames per second (quite good, because it's pal). I also tested ffdshow(about 15 fps, postprocessing off) and divx 5.00 (about 12 fps, postprocessing off). The film was watchable only with sigmas decoder.

[NeVeRLiFt]

http://63.231.44.178/board/showthread.php?s=&threadid=33624272

go there to see the details

I unistalled this codec and now everything works, no more problems.


Special Note: The tests I did showed it has good quality, I did not test for size vs quality... I have removed this codec and want use it till the problem is solved with DivX5 and XviD playback. With this Sigma Mpeg4 codec installed XviD was messed up but worked, and DivX5 would not work and caused the video to freeze and lockup my pc. Unistalled Sigmas codec and no problems.

Hope there is away to fix this.


The link may have been down so I'm adding the info from the post here.

Sigma REALmagic MPEG-4 Video Codec
http://www.sigmadesigns.com/product...deo_encoder.htm

Special Note: when selecting the AVI FourCC... the RMP4 want work with Virtualdub, that means no compressing the Audio.

The other 2 FourCC work and are smaller in size but the quality of the Video is sorta not as good cause there is no post process filtering.
Also when seeking I noticed the video gets messed up and then gets clear again.
Best bet is to just play around with it and see what you think of the quality. I think the Quality is very good and the size is good to.
This is VideoCapture I'm talking about here not Encoding.....

I guess we may be seeing alot of Mpeg4 codecs popping up


I know with the little bit of testing I have done with this codec it can work for VideoCapture. But so can DivX5.02 Pro, and XviD and even the older DivX3.11a


Do not use this!

Unistall now and there will be no problems.
If you try to use it... it will mess up the DivX5 and XviD codecs during playback.
DivX5 will even lock-up and crash!
Unistall the Sigma Mpeg4 codec and everything is fine.
You have been warned!

Trust me just use DivX5 and keep this codec put up some where.

Sorry about this.
Also if you have problems with DivX5 and playback go here and read.

http://forum.doom9.org/showthread.php?s=&threadid=19404

In the first post is the answer.





http://forum.doom9.org/showthread.p...6816#post146816

Maybe keep your eye on this post. See if they can find away to work around it and make everything play nice together

[Dali Lama]

Originally posted by Latexxx
This codec has fastest mpeg-4 decoder that I have ever seen.

This is definitely true. On a Celeron 733 I can playback 320x240 @ 59 fps! Yes 59 fps. This is very good news for those decoder lovers out there.

Originally posted by TactX
When searching through the AVI in MediaPlayer (also VDub) I got real bad artefacts until the next I-frame !


Yes this is also true form my testing. Basically its similar artifacting to Divx 3.11 smearing effect. This happens mainly if I seek backwards and as TactX said lasts until the next keyframe. This is a little bit of a bummer.

Originally posted by TactX
REALmagic is a bit faster (~9%)


First, I would say 9% is a lot faster, especially for full-length movies. Second, I got equivalent speed increases using B-frames.

A movie clip encoded at exactly the same settings except I included B-frames for REALmagic:

3min 50s for Xvid compared to 3min 10s for REALmagic.

That's ~8% faster.

Originally posted by TactX
Using Q=4 it looked worse than XviD (Q=3) on "Neo vs. Morpheus fighting-scene" and the file was even bigger ! (XviD-decoding by ffdshow but all options unchecked)


Ok this is where it gets really interesting. The quality/file size issue is of most importance.

First, I don't think comparing codecs by quantizer, especially only one quantizer value is accurate. I chose to do 100 percent quality comparisons instead. Since I just started testing, let me just give some preliminary results.

Basically under the same settings REALmagic produced a file of approximately half the filesize of Xvid at 100% quality settings.

This is really astonishing. I will continue to test and see if I can replicate the results. I encourage all to do the same.

The only difference was that REALmagic used B-frames.

I did not limit the quant values (1-31) and I used both codecs psychowhatever features.

As I remember Xvid produced a filesize of approximately 44MB and REALmagic produced a filesize of approximately 22MB.

A BIG HOWEVER:

To my eyes the visual quality of Xvid @ 100% was a little better than REALmagics @ 100%. Possibly due to the use of B-frames again.

I really should do other tests without B-frames in REALmagic to see if results differ.

Just wanted to inform everyone who asked for results.

Dali

[Dali Lama]

Test Results Updated:

Video Content:
Progressive DV
Style - High Action

--------------------------------------------------------------

Settings Xvid:

Encoding Mode - 1-Pass Quality
Quality - 100

Motion Search Precision - 6 (Ultra High)
Quantization Type - MPEG
FourCC Used - Xvid
Maximum I-frame Interval - 300
Minimum I-frame Interval - 5
Enable Lumi-Masking - No
Enable Interlacing - No

Min I-frame Quantizer - 1
Max I-frame Quantizer - 31
Min P-frame Quantizer - 1
Max P-frame Quantizer - 31

Settings Sigma REALmagic: #1

Rate Control (RC) Mode - Variable Bitrate (Quality)
Quality - 100

Bidirectional Prediction - No
Adaptive Quantizatoin - No
Intelligent Deinterlacing - No

Interlaced Video Coding - No

Select AVI FourCC - RMP4

Minimum Quantizer - 1
Maximum Quantizer - 31

Quantization Tools - MPEG

Maximum Keyframe Interval - 300

Settings Sigma REALmagic: #2

Rate Control (RC) Mode - Variable Bitrate (Quality)
Quality - 100

Bidirectional Prediction - Yes
Adaptive Quantizatoin - No
Intelligent Deinterlacing - No

Interlaced Video Coding - No

Select AVI FourCC - RMP4

Minimum Quantizer - 1
Maximum Quantizer - 31

Quantization Tools - MPEG

Maximum Keyframe Interval - 300

--------------------------------------------------------------

Xvid vs Sigma #1

Filesize | Perceived Quality | Time to Encode

Xvid 49.2 MB | A+ | 4min 47sec

Sigma #1 26.0 MB | A | 4min 31sec


Xvid vs Sigma #2

Filesize | Perceived Quality | Time to Encode

Xvid 49.2 MB | A+ | 4min 47sec

Sigma #2 20.9 MB | A- | 4min 33sec

--------------------------------------------------------------

Difference Between A+,A-,A is very little in comparison to clean source material.

Note how little the encoding times differ between Sigma #1 and #2.

Also This is the second time I have achieved drastic differences in filesize between Xvid and Sigman. We are looking at ~50% reduction without B-frames and ~60% reduction with B-frames.

--------------------------------------------------------------

I decided not to test the psychovisual whatchamacallit features because they tend to bring up controversy regarding "ugly" frames and if the result really looks better or not.

Just to give thoughts on Sigma's "Adaptive Quantizatoin" feature. It performs well to my eyes. Something similar to a light Divx5 psychovisual.

--------------------------------------------------------------

Well thats it for me now. I hope I designed a fair and accurate test for both codecs. I'm interested in others opinions of this codec too.

Take Care,

Dali

[TactX]

This is strange.
I've done a 2nd test this time with 100% quality set. Again "Neo vs. Morpheus fighting-scene".

Same setings for both except:

- XviD with luma masking

- REALmagic with B-Frames and Adaptive Quantization.

Result:

XviD: 57674 kb
REAL: 60034 kb

Quality was pretty much the same (though I would say that XviD looks a bit better).

[CavalloPazzo]

I've done some test too, and I founded that with sigma quality 100% = constant quantizer=2 (even if minumum quantizer is set to 1; Even encoding with constant quantizer =1 or 2 give the same output). So with constant quantizer=2 in Xvid too, xvid files is a little smaller. If you set 100% quality in XVID quantizer 1 is used, so the file is bigger then sigma's one.
With some test (very short clip) file with bframe are bigger then files without them. Dunno why but my results are different then yours. Maybe I should try with longer clips.

[TactX]

Well, I've encoded a complete movie "Memento" as 1-CD.

- XviD was done using above settings and 2-pass

- REALmagic with above settings and quality setting 70%

Result (both without end-credits):

- XviD: 596716 kb

- REAL: 621022 kb

And STILL XviD does look better !

[Stabmaster-Arson]

thats good, i was gonna ask about 2 pass. For encoding movies I have absolutely no use for any 1 pass codec anyways.

Dali lama, any hance you could encode those same clips again with Xvid using 2 pass and see the difference ?

[TheXung]

ok everyone, take a few minutes to go to the bathroom or shit in your pants because we have a new god. Yes this codec is the shits. From what I can tell, it's does variable bframe intervals and it's fast like anything. Apparently it is not equal to compare a clip at a specific quant between the codecs because this one if off from the others.

Anyway, these are some shots, magnified for the quality impaired:
RealMagic, quality 75, filesize 20.2MB, bframes on, adapt off, MPEG quantizer
http://www.people.virginia.edu/~xn4f/R1.jpg
Xvid, quant 3, filesize 20.1, bframes on (interval 1), lumi off, MPEG quantizer
http://www.people.virginia.edu/~xn4f/X1.jpg
Divx 5.02, quant 3, filesize 19.2MB, bframes on, GMC on, 1/4 pel on
http://www.people.virginia.edu/~xn4f/D1.jpg

Notice how much more detail there is in the grass, shield, and horse's head plate. From the looks of the level of detail, it appears that RealMagic has implemented 1/4 pel motion estimation even though it's speed suggests not.

These are from the same clip, just different shot
RealM
http://www.people.virginia.edu/~xn4f/R2.jpg
Xvid
http://www.people.virginia.edu/~xn4f/X2.jpg
DivX
http://www.people.virginia.edu/~xn4f/D2.jpg

Again notice how much more detail there is in the guy's red shirt and hair.

For those testing this codec, it appears that it's adaptive quantizing, rather than increasing the quants in areas less noticed, it decreases the quants in areas that are more demanding. So at a specific quant, adaptive quantizing produced a larger but better looking video.

Another note, perhaps if the DivX rip had been given the extra megabyte, it would have equaled the quality of the RealMagic. But DivX is still twice as slow.

This codec is definitely up there.

[NeVeRLiFt]

Anyone else have this problem? With the Sigma Mpeg4 codec installed DivX5 want playback and XviD is sorta messed up to.
DivX3.11a plays fine. I unistalled the Sigma codec and everything works like it should.

[Selur]

why I don't 'like' the codec the way it is atm:

- macroblock's if u use b-frames while encoing and if u scroll (only till next I-frame)
- not so good performance at lower datarates (xvid and divx perfrom way better)
- sometime there's some kind of halo around objects if they are in front of a dark backround (e.g. spaceships in outer space)

All in all I have to say that the codec looks promising not as good as DivX or Xvid, but I think it's worth to keep an eye on this codec.

Cu Selur

[temporance]

I agree Selur - i don't like the 'feel' of the video made by this codec either. However much I try, it never performs as well as DivX 5.0 Pro, + Bframes.

Maybe Sigma's next release will be better? Let's hope so.

[Acaila]

My first impressions:

Settings:
REALmagic
1-pass 100%
B-frames
Adapt. Quant.
MPEG

XviD:
1-pass 100%
Lumi Masking
MPEG

DivX 5.02:
1-pass 100%
GMC
B-frames
Q-Pel

Quality:
REALmagic seems almost identical to XviD, although there is obvious ringing around sharp lines which XviD has much much less. DivX5 has much less detail (and noise :)) than either of the other two.

Filesize:
REALmagic -> 28.214 kb
XviD -> 29.874 kb
DivX 5.02 -> 23.792 kb

Speed (3080 frames):
REALmagic -> 50" (62fps)
XviD -> 57" (54fps)
DivX 5.02 -> 1'44" (30fps)

So that's a nice boost :)

FOURCC:
RMP4:
Encodes -> ok
Plays -> ok
Open encoded movie in VDub -> no can do (VDub requires VfW decoder, which RMP4 doesn't have)

DIVX:
Encodes -> ok
Plays -> very choppy
Open encoded movie in VDub -> frames are all garbled

DX50:
Encodes -> ok
Plays -> very choppy
Open encoded movie in Vdub -> ok

I admit this was just a single test with a movie trailer, maybe I'll compare an entire movie this weekend.

@Neverlift:
I don't have any problems with XviD or DivX5.02 playback, everything seems to be a-ok.

[Latexxx]

The results are similar to what I've got whit it.

[Chibi Jasmin]

Thanx for all your results :-) I won't install it at the moment to be on the safe side, but let's keep an eye on this one...

[Acaila]

Just finished doing an entire movie with it (Bitrate=976, B-frames, Adapt.Quant.). The quality really sucks. The fact that it only has 1-pass modes shows, macroblocks aplenty on high motion frames, the low motion scenes look great though.
It also came out approx. 20MB undersized :angry:

Conclusion: It looks like this codec has potential, but I wouldn't touch it again until it has at least 2-pass VBR encoding, maybe it could be useful for direct capture encoding until then.

[Latexxx]

Has anubody tried using the decoder with divx or xvid ?

[TheXung]

I don't see what you guys have against no 2-pass mode. Just use it on quality based. It's not like you get that much more from using 2-pass over quality based.

[Zeus]

I compared the brandnew RMP4 v1.0 with the DivX v5.02 Basic software codec, because both of them are freely available and have no ads included.

The test-sequence is a sample of "Shrek" (the popping bird scene) with a very high amount of details and a very clean picture.

Test #1
-------

goal: maximum visual quality, variable: file size (bitrate)

RMP4 settings: VBR (Quality)=100, bidirectional=on, adaptive quantization=on, Quantization by MPEG, keyframe intervall=100

DivX settings: 1-pass (Quality)=100, psychovisual=strong, keyframe intervall=100, Performance/Quality=slowest

results:
DivX: visual quality: very good, file size: 77221 KB
RMP4: visual quality: very good, file size: 74805 KB (-3%)

conclusion: the visual quality is nearly the same, but the smaller filesize of RMP4 is very promising...

Test #2
-------

goal: fixed bitrate @ 700 kbps, variable: visual quality

RMP4 settings: VBR (Size)=700, bidirectional=on, adaptive quantization=on, Quantization by H.263, keyframe intervall=100

DivX settings: 2-pass=700, psychovisual=strong, keyframe intervall=100, Performance/Quality=slowest

results:
DivX: bitrate: 689 kbps, visual quality: D
RMP4: bitrate: 708 kbps, visual quality: C

conclusion: while reaching the desired bitrate, the visual quality of RMP4 is much better than DivX.

A last note to all of you missing the settings for post-processing: all the options are turned off. Why? Ask yourself: Do you think, any future hardware player gives you post-processing of the video like DivX codec? I think not...

greets,
Zeus.

[Teegedeck]

Nic is currently disassembling RealMagic to compare the code against XviD's. Still too early to say something with certainty but I'm afraid some hints indicate that the code is "very similar". I hope this turns out to be purely accidental.

[Nic]

Unfortunatly, very similar appears to be an understatement, but ive posted at Xvid.org & hopefully the main developers will delve in too.

-Nic

[Koepi]

@Nic,

no need to wait, they already did...

I'll forward you 2 mails if you like, PM me your eMail-address again pls.

Regards,
Koepi

[NeVeRLiFt]

Man I dont know what to think about them using XviD!
This Sigma codec seems to do pretty good even though its only 1pass.
And also tested it for VideoCapture and it does well.
But I have a problem.... when I install Sigma's codec DivX5 encoded movies want play. And XviD losses it playback filter even though it still plays.

I have done the Reg tweak so DivX5 does not take over and playback every thing.


Maybe the good stuff can be taken from the Sigma codec and used in XviD ;)

[gldblade]

>I'll forward you 2 mails if you like

My guess, don't expect a reply to those e-mails...

>Maybe the good stuff can be taken from the Sigma codec and used in XviD

Jeesh, that would be tough. Although the above speed reports are great...

[Koepi]

Use search precision 5 and redo the tests - there you go with your speed boost.

Regards,
Koepi

[int 21h]

http://forum.doom9.org/showthread.php?s=&threadid=28210 is probably the place to discuss any XviD related issues so that we can have it all condensed into one place.

[Koepi]

Please, stop discussing this for some days.
We don't want a pre-biased public or users, so just cope with it for now - you're doing bad things to the XviD developers by talking bad about sigma.

If some supermod could close this thread now it would help us really much!
I hope you'll understand that.

Just wait for news from us in this matter!

Thanks,

best regards,
Koepi