My new codec

[fefalcon]

http://www.libreriamondadori.grosseto.it/go!eye.htm

This is only one shot. When it will be ready, you'll have the software.
Be mind: this is not for DVD-copy

[Neo Neko]

Hmmm interesting. Does not pertain to much to the board. But interesting. You say this is a codec, but is it integrated into the program shown or is it independant. Is this a new work or is it based off Momusys/opendivx/Xvid etc?

[fefalcon]

It's based on free source ISO 14496, the only one object based. The next step is to make it quicker in coding (everyone knows ISO limits i suppose) , on the other hand for it's final use (cif-qcif format, extremely low bitrate, mobile oriented) it's ok.
A lot of changes in ISO code are made to be possibile the integration of:
1) RTP streaming (excellent results in local network and p2p adsl, some bug to be fixed in other case like adsl-56k or 56k-adsl)
2) Intelligent codec: the codec looks at the "world around", He ( ;)) understands the scene, the objects, the statistical movements and makes the excellent setting.
3) In my application i suppose a fix telecamera (as said in overseeing application)so the static background has a wavelet coding (like jpeg2000 you know). Objects still (but no forever...) in DCT.
4) Psychomask based on visual brain theory (i.e. HVS, human visual system.) In simple words the codec studies where eye couldn't see the noise. (but it's a total different approch from divx psychomask or Xvid lumi masking). Excellent results in bitrate (max 40% minus), with MITRE IQM (quality measurement) i loose only 2-3% in relative quality (but also with the usual PSNR i loose max 2 db).

Be sure when it's ready, i'll give you the software for a test.
sorry if it's off topic, but i didn't understand where to post. In effect i see most of all interested in frame based coding (that is dvd-copy)and sure for this divx it's the best.
But i hope evereyone doesn't forget the real soul of mpeg-4 theory, the objects, and Projects Mayo (or other) doesn't deal (or not interested) with this.

[Razorblade2000]

Be sure when it's ready, i'll give you the software for a test
Any predictions when this will be?
Will there be a beta?

[Neo Neko]

Originally posted by fefalcon
sorry if it's off topic, but i didn't understand where to post. In effect i see most of all interested in frame based coding (that is dvd-copy)and sure for this divx it's the best.

You are correct that most everything here pertains to DVD conversion. But still this is intriguing. I think it would be something we would all like to see.

Are you doing this as a hobbie or with goals for business in mind? If you are not doing this in conjunction with/for business you might considder contributing to or cooperating with the Xvid group. Your project your choice of course. But it would be nice to have a more complete MPEG4 codec rather than another specialised one. Not that it would not be usefull. :)

[fefalcon]

Originally posted by Neo Neko

You are correct that most everything here pertains to DVD conversion.

Most everything? Well, best said totally everything... Maybe MPEG4IP does something more interesting, but it crashes with objects.

Originally posted by Neo Neko

Are you doing this as a hobbie or with goals for business in mind?


Well, for the moment it's my thesis opera ;) and this version it's enough for it. Surely not for business if there will not be anyone interested in it. Maybe for hobby if I'll find time in future.
I couldn't say now.
In october I'll publish a page with a powerpoint and pdf explanation of this project.

Originally posted by Neo Neko

If you are not doing this in conjunction with/for business you might considder contributing to or cooperating with the Xvid group. Your project your choice of course.

Just done in the past, the answer was they are interested to obtain DVD quality only. As you easily could understand, my project is the complete opposite. For overseeing targets (and in this moment, i think, the security is an important point, damn, MPEG4 is really only for DVD copy?!?) is not important an excellent quality, but a great compromise with bitrate it's the final purpose. (think about a multicasting with 10-20 camera/server and one client ...)

The actual problem is only to obtain a quick dynamic segmentator (in my project i create a static segmentator for the static background) to extract objects from the scene.

[Neo Neko]

:( While what you put forth would have no direct effect on what they considder to be DVD quality. I could see where it could be the start of something that would. Oh well. Keep us updated. :D

[fefalcon]

Here is final graphic release.
In october i think the software will be ready.

[MfA]

DVD quality and broadcast quality coding of video is interesting to a community, security cam's are interesting to business ... some things are just not suited to community based open source projects.

[[Toff]]

Would this type of codec be suited for video conference ? It could be great as it use RTP.

Or maybe you can use it to watch over you children, your wife, your house or your red fish with a simple webcam.

[fefalcon]

DVD quality and broadcast quality coding of video is interesting to a community, security cam's are interesting to business ...
Maybe, but i hope community interests are not limited in DVD copy. I'm talking about MPEG4, the divx-xvid and others are only raped versions of the standard.

Would this type of codec be suited for video conference ?
Yes sure.

Or maybe you can use it to watch over you children, your wife, your house or your red fish with a simple webcam.
well it's a bit more complicated... ;) Maybe because in community immagination there is not the sense of object power. In the shots you could see the reconstructed frame, but it's really more interesting than a simple webcam. When you have objects you could do a lot of interesting things. About psychovisual mask/intellingent autosetting is a way to obtian best quality with lowest bitrate.

[-h]

Maybe, but i hope community interests are not limited in DVD copy. I'm talking about MPEG4, the divx-xvid and others are only raped versions of the standard.

I can assure you that none of the XviD developers have raping the standard on their todo list.

-h

[int 21h]

Originally posted by -h
Maybe, but i hope community interests are not limited in DVD copy. I'm talking about MPEG4, the divx-xvid and others are only raped versions of the standard.

I can assure you that none of the XviD developers have raping the standard on their todo list.

-h

Yea.. just look at the latest CHANGELOG

11/09/02 - Finished raping the MPEG-4 standard, guess we'll move on to raping something else, maybe we can use some of the descriptors in MPEG-7?

See, they've already moved it out of the todo list! Anyways, MPEG-4 defines a video standard, video can be used for lots of things. Especially porn. If I start a porn website, encode all my porns to full resolution XviD, does that mean I too am misusing the Mpeg-4 standard?

Of course not. I think you underestimate what great lengths to XviD team has gone to, and the particular attention to detail they have in regards to 'the standard'.

[fefalcon]

I can assure you that none of the XviD developers have raping the standard on their todo list.

Well, maybe what i meaned it's a bit off topic. The code developed by Divx ot Xvid team is anyway called MPEG-4. In my view this is totally false. I think it's a really good way to copy a film frame by frame (and if these frames are porn, friends, animals or news it's another problem). We can call them the MPEG-2 Successor, MPEG-3,MPEG-5 or whatever you want. This is NOT MPEG-4 soul. Mpeg-4 is object based, but it's so difficult to working on object based. And frame based codec is a raped version of the standard. There is no MPEG-7 and MPEG-21 until MPEG4 is really as it would be.
Anyway, for their purpose (be onest, only dvd-copy, we are still far from dvd-streaming)and maybe hw version, these codecs are surely great things, and i hope Xvid could reopen soon its work.

[Koepi]

Can you please cite the MPEG4 Simple Profile specifications?

I want to see specially the "object only" part of it, thanks.

Koepi

[fefalcon]

You know the answer. Mpeg-4 profile standard is made to create everthing, to use it in a lot of way.. maybe someday we'll see MPEG-coffee.
But maybe the question is another, how difficult is to create a real good segmentator? Or, are objects useful for DVD-copy?
I really hope someday we'll see something really innovative and interesting in other direction.
But i have just confirmed that you-Xvid and Divx are EXCELLENT for frame purpose and I always express my preference for xvid like an open source without boring spyware or everything else.
On the other way, my conception of the mp4 is different, i really think MPEG-4 power is in masking, but because of wide interpretations, it's only my opinion.
Just last consideration. Someone has said comunity is interested expecially in DVD-copy. Yes at this moment it's true, it's sointeresting to have some good transcoder from MPEG-2 to put a DVD in 700M in 6-8 hours. But what will happen in future when DVD recorder prices will be low, or with blue laser tech. on cd? I think things will be different.

[-h]

But maybe the question is another, how difficult is to create a real good segmentator? Or, are objects useful for DVD-copy?
I really hope someday we'll see something really innovative and interesting in other direction.

Sure, the XviD team only really looks at parts of the standard that might be useful for backing up DVDs, or similar frame-based pursuits. I don't think anyone has the intention of implementing the full MPEG-4 standard (the whole mammoth thing), at some point there has to be a cut-off between what's actually useful to 95% of people and what would take thousands more man-hours to code.

In that respect XviD is an easier project to work on. Your work is way out in the "frontier" I would say, and may well bring quality benefits far beyond what we've been able to achieve. This is good because I'm pretty sure no one else has been driven to try it.

-h

[fefalcon]

I totally agree. On the other hand the problem to move my project to a transcoder Xvid like, is that actually there isn't a segmentator able to work on moving camera (as in a movie). But anything is impossible... ;)
On the other hand i'm sure that my psychomask could surely bring benefits on xvid luma masking. But this is made for still image in wavelet coding so we have (finally....) to go away from DCT and its horrible blocks. My first psychomask is on block 8x8 based, on the other hand we could transform it in a different tree-correlation-pixel based, why not.
About quality? There is always the same compromise between quality and bitrate (quantization rules... but as luma masking did, and psychomask does in a different way, we have to find the right way to put our noise... ;) ). When i could i'll publish results obtained by IQM-MITRE (not by the 'stupid' PSNR).
Thank you for attention.

[Defiler]

Stepping aside from the mini-flamewar that almost developed a bit earlier in this thread, I'd like to say a few words..

What we have here is another example of a very interesting phenomenon that seems almost unique to the "desktop video" world.
"Hi, I'm new here. I just created this amazing thing that no one thought could be done; by myself, in my spare time. Comments, please?"

At least from my perspective as an end-user, rather than a coder, this seems to happen with amazing frequency. I love it, and I can't think of any good examples of this happening in other fields of software.. at least not recently.

I for one will be glad to see the death of DVD, and with it, the innumerable restrictions imposed upon us in terms of colorspace, resolution, framerate, etc. Imagine owning a digital camcorder that shot direct to full-on MPEG-4, without chopping out half of the color, and 90% of the resolution? Think of what independent film makers could do with that?

fefalcon: Good luck with your thesis, and I look foward to testing out the software you have offered to us.

[fefalcon]

Thanks a lot Delfier. Again i also precise that my intent is not to denigrate other intents (expecially when they are free source), only to express my opinion on mp4 world, in my point of view, actually too much focalized on DVD copy without any idea toward other horizons. This is my only disappoint, nothing else.

the innumerable restrictions imposed upon us in terms of colorspace, resolution, framerate, etc. Imagine owning a digital camcorder that shot direct to full-on MPEG-4, without chopping out half of the color, and 90% of the resolution?
First point:
Well, in reality when we talk about compression something we have to cut off, by force of circumstances. My intent is to cut off where eye couldn't see, on the other hand it is stupid to leave something that eye, or visual brain (on which is based my psychomask), cannot perceive. As i said before, for storage purpose, this can be got over when blue laser will be in our home at low price. But what about streaming? Yes, broadband is now a strenght reality, but still very limited. also think about a server that has to manage 10-30 client camera....
Second point:
I thought in my thesis that is very boring to set all parameters to obtain the best results for the scene, also because an inexpert user could be scared in front of mpeg-4 settings. So i create a way to make and intelligent codec, that sees, understands and learns.
Third point:
Again, objects, again. This is my passion ;). When there will be a real good dynamic segmentator there will be a lot of new improvements, you couldn't imagine how many. Just an example: in the server-client camera said before, you could create a virtual world to manage all the object (people, car...) and you could avoid to look at 10-30 monitor at the same time. (and spend less money for the oculist!! :D ).

[Defiler]

Imagine having a camera linked to a system that could accurately recognize objects it found in a database. Now mount it in a pair of sunglasses. Want to know what the resale value of a 2000 Honda Accord is? Just look at it. Heh.

Anyway; will your thesis be available online when it is finished?

[fefalcon]

Yes sure! Great! In fact other purposes are face and car plate detect.
After my thesis in October I think I'll publish a powerpoint file and the software. I couldn't give it for sure, it depends on my prof. plans (i've heard about IEEE publication... but it's not sure yet).

Just last thing. Sorry if I troble you again and I hope moderators could excuse me if this post is a bit off topic.
I simply need your help. After 'stupid'PSNR and Mitre-IQM, I'd like also to taste quality by really human impression.
I'd like that you partecipate in my pool, to evaluate the parameter quantization-quality and psychomask level.
You have to find the picture psychomask-derived, generally at less objective quality, but with great gain in bitrate.
Some pic is easy to find, but other... ;)
But you have to look at them only few seconds... don't cheat!!:o
http://www.libreriamondadori.grosseto.it/test1.htm
I hope to find just 100 kind guinea pigs...:p
Thank to all for your time

[Defiler]

I took the test. I think I did well. :)

[Gabriel_Bouvigne]

I this test I got the same results as most of the people, except for the last one. I am curious to know the real answers

[Gabriel_Bouvigne]

Btw, did you checked this:
http://forum.doom9.org/showthread.php?s=&threadid=19705&highlight=image+AND+comparison

[fefalcon]

mmm the test is going under my aspectative. Even if i can't control if you really look at them for few seconds.

I put this BTW:
NB: You have to give a look at pictures only few seconds, without looking for small differences. This test is made in the sense ' at first sight'

and for someone who hasn't understood anything about masking:
in simple word you must choose which in your opinion is the worst.

Thank you again.

[fefalcon]

Some of you has asked me by email why the strenght rule "at first sight".
Simply because Psychomask is intended not for still image, but it is integrated in the codec and it is also based on psycho-tracking. In other words it means that after a quick eye-update about the world around, your brain-eye will focuse on the object and it's very difficult you'll be able to see fine details.
Thank you.