DivX Pro Codec - Fusion Beta Release

[IgorC]

problem with YV12 codec (fusion pack)
image of *.avs was rotated 180º. there is no problem in Vdub, but when playing *.avs in MPC

and *.avs in Recode2 (Nero AVC) ,etc. So I had to uninstalled fusion codec. Before I had

Helix YV12 codecs.

fusion version.
some minor visual impovements. but video is still so smooth, maybe less visual artefacts but

it´s mean less details. It would be good if divx 6.0 will keep details as Xvid or even better.(I undesrtand that´s topic is FUSION BETA)
700 kbits
Xvid 1.1 (2pass): max. quality options + custom Matrix EQM ULR rev 2
http://img94.exs.cx/my.php?loc=img94&image=2passxvid700ulr1ln.jpg

Divx Fusion (3pass.each insane Q.) : MPEG cuant (I tried to keep the details).
http://img108.exs.cx/my.php?loc=img108&image=3passdivxfus700nopsycompeg3ri.jpg

Source : http://img92.exs.cx/my.php?loc=img92&image=avsoriginal640x4804ht.jpg

the same result in test with options for DivX devices.


is there any need to 3d pass? will 2 passes at insane quality be enough for 98-99 % possible quality.

[gldblade]

I'll assume you've mixed the Y and X axis up, because as it stands it makes absolutely no sense.

It just shows steady increase of PSNR at different bitrates.

Isn't it supposted to show PSNR at a steady increase of bitrate? The graph is indeed very confusing, but its right. What its showing is that different DivX modes requires different bitrates to achieve the same PSNR. For example, in order to reach the PSNR of 50 db in clip Q01, DivX Fusion Insane requires ~88 kbps, while DivX 5.2.1 Fast requires ~106 kbps. As you can see, different DivX modes require different bitrates to achieve the same PSNR.

What doesn't make sense about the graph is that they decide to connect the dots between the different clips. You can't interpolate between the dots like that, what would such an interpolation mean? This connection of the dots leads to confusing conclusions. For example, examining the curve for XviD Motion 6 Trellis On, the graph implies that it takes a smaller bitrate to acheive a PSNR of 50 db than to acheive a PSNR of 37 db.

[stephanV]

Originally posted by gldblade
The graph is indeed very confusing, but its right. What its showing is that different DivX modes requires different bitrates to achieve the same PSNR. For example, in order to reach the PSNR of 50 db in clip Q01, DivX Fusion Insane requires ~88 kbps, while DivX 5.2.1 Fast requires ~106 kbps. As you can see, different DivX modes require different bitrates to achieve the same PSNR.

The bitrates there are relative, not absolute. So, according to the graph, DivX in extreme mode, Q01, needs about 90% of the bit rate Xvid with no trellis on uses. DivX balanced needs about 102% of XviDs bitrate.


What doesn't make sense about the graph is that they decide to connect the dots between the different clips. You can't interpolate between the dots like that, what would such an interpolation mean? This connection of the dots leads to confusing conclusions. For example, examining the curve for XviD Motion 6 Trellis On, the graph implies that it takes a smaller bitrate to acheive a PSNR of 50 db than to acheive a PSNR of 37 db.

Again it is a relative bitrate not an absolute one. The Q value does not stand for different clips, but for the constant quant value used in the encoding. Therefor some interpolation is allowed. There is however an error in the interpolation and is that they have decided to make a smooth line out of it, while there seems to be no mathematical principle for the correlation. It is very difficult to predict what happens between for example point Q=3 and Q=2. In such cases drawing straight lines between the points is better as you only can determine the general trend of the line. Not how it actually is.

[nicco]

SMD
That's not the point I've ever been trying to make!

A couple of years ago I decided to see "how low you could go" at generating Mpeg4 in SP mode using full DVD resolutions.


Ok!:)

[dragongodz]

DVDx 2.3 and Tmpgenc 2.5(avi output) both can now set settings with this version. the last beta they couldnt of course.

quick 1 pass tests.
Tmpgenc, feedback window disabled, encodes fine.
DVDx, feedback window disabled, encodes fine but takes nearly 30 seconds to start encoding. other codecs including older Divx 5 versions do not have this problem.

enable feedback window will crash both Tmpgenc and DVDx.

tested on Athlon XP 2400+, Windows XP SP1.

[billou2k]

I've done some tests with Fusion and interlaced content but it looks broken for me. (progressive content works ok). The encoding process works fine (looking at the expanded feedback window, everything looks fine) the problem just happens during the decoding.
The errors on playback look like blocks are not refreshed properly. I'm using the DivX Decoder not ffdshow and virtual dub confirms the decompressor is DivX Fusion.
Did anyone notice the same problem?

[DigitAl56K]

Originally posted by dragongodz
DVDx 2.3 and Tmpgenc 2.5(avi output) both can now set settings with this version. the last beta they couldnt of course.

enable feedback window will crash both Tmpgenc and DVDx.

tested on Athlon XP 2400+, Windows XP SP1.

Thanks for the report.

I can't duplicate this on TMPGEnc-2.524.63.181-Free, the Feedback window works for me. Did you configure the codec form TMPGEnc before you began encoding the first time?

I can duplicate with DVDx, but only when YV12 colorspace is used. If I switch to RGB everything works perfectly. I am suspicious that this is actually a defect in DVDx, because I remember a similar problem being reported over at the DivX.com forums some time back where the solution was to switch the output to RGB mode. I'll log it in our QA system and see what our engineers make of it.

[dragongodz]

TMPGEnc-2.524.63.181-Free, the Feedback window works for me. Did you configure the codec form TMPGEnc before you began encoding the first time?
i used 2.521 so i will update to the latest version later and check against that aswell. and of course i configured the codec first.

I can duplicate with DVDx, but only when YV12 colorspace is used. If I switch to RGB everything works perfectly
YV12 is broken and no codec will work with it, thats been known for a long time. i tested with both RGB and YUY2 which work with Divx 5.0 to 5.21 fine.

I am suspicious that this is actually a defect in DVDx
no offense but thats pure bullsh*t. the fact that DVDx can use Divx 5.0-5.21, Xvid, VP6 etc etc etc fine should tell you it is not DVDx thats the problem.

I'll log it in our QA system and see what our engineers make of it.
you may want to tell them to look at the dvdx/AVI_LIB/avi_lib.cpp file.

oh and you may want to look at the names in the DVDx window when you start it. :D

[dragongodz]

first i think i know the 30 second delay problem. the data for that is coming off a crappy old work/data hd while the other tests were froma newer/better hd. the old hd is showing signs of age and some things are copying full speed while others are going at varied speed.

the real issue however is another story. i downloaded the Tmpgenc version you mentioned and it suffers the exact same problem. encodes fine with disabled feedback window, freezes up with feedback window enabled. i even got around to testing it with virtualdub 1.5.10 and same thing. yes freeze with feedback window enabled.

so it is definatly the feedback window my system does not like since disabling this enabled all the programs to encode. it does not even start to be drawn etc before freezing the application(and infact slowing windows down to a crawl in the process).

so some system info which may or may not help.

Athlon XP 2400+, 256mb ddr ram
Windows XP SP1, directx 9.0c
installed software includes, visual studio 6 + SP5, directx sdk 8.1, windows sdk, intel c++ compiler 8.0

let me know if there is anything else you want to know or if you want me to test a debug version etc.

[SeeMoreDigital]

DivX guys,

Can any of you confirm why your "direct-show" decoder filter (v5.3.0.1453) has soared up in size to 781KB?


Cheers

[AsTimeGoesBy]

Interesting posts here!
So i also like to make some test samples so see how DivX Pro Fusion Codec (http://labs.divx.com/archives/000045.html) works for my belongs comapred to the recently released Xvid 1.1b (http://www.koepi.org/xvid.shtml).
However i have so me small questions i haven't found any answer by searching the web. - Would be nice if somebody could please spend some ssconds to post some answers.

- Psychvisual Effects: Masking mainly only for Anime or Comics, right!? I'm sure i have readt about this anywhere but i wasn't able to find it again....:(

- After installing the Divx Fuison pac kI couldn't find any configuration tab/window to control the post-processing level. - Is there any at all?
As i have experienced till now i guess there is any post-processing on DivX files but i'm not sure how and if that also woarks on Xvid files.

- In VirtualDubMod the 'DivX YV12 codec' is not available!? - Wouldn't that improve speed when using AviSynth?

[billou2k]

Sorry to bother you again with that
but it seems that the encoding of interlaced content is messed up in Fusion.
All the decoders have tried show the same faulty video (misplaced blocks)
Nobody got any info on that?

[LordIntruder]

Originally posted by AsTimeGoesBy

- In VirtualDubMod the 'DivX YV12 codec' is not available!? - Wouldn't that improve speed when using AviSynth?

Forgot to mention that last time. I also noticed that immediately within VDM. There are two lines:

- Divx Fusion Codec
- Divx Fusion YV12 Codec

The second choice give me an error in the format restrictions window at the right. I'm not alone to get that, good point :D .

--

billou2k: I assume you use the interlace option within the codec? I have just make a test on an interlaced DVD source using a deinterlace filter and avisynth and the result is perfect. Provide more details on what you have done, what settings you choose and if you use an AVS, copy paste it here.

[stephanV]

The YV12 codec is for decoding YV12, it does not encode to DivX or anything :)

[MacAddict]

Before I install this on my HTPC...can anyone tell me if the settings dialog is actually resizeable now? On lower resolution screens the dialog box is massive in length and the buttons at the bottom are cut off.

[LordRPI]

Doesn't look like it is yet :(

[billou2k]

I'm encoding some shorts sequences from TV, I'm storing them as YUV uncompressed AVIs. They are PAL (720x576) pure interlaced and they appear properly in VDUB.
I set up Divx Fusion (Build1453) as
unconstrained,
Extreme quality
Adaptive multiple cons. B frames (but I tried single and no B frames)
no GMC
noise red. off
psycho enh. off
interlace: preserve TFF
h.263 (tried h.263 optimised)
1 pass 2000 kbps

And this gives blocks moving all around and producing really weird effects...

Oh and I've made some progress!
Changing the codec performance to "Balanced" solved the problem!
And "Better quality" causes the same problems as "Extreme quality".

What's the difference between balance and Better quality and extreme quality that could cause such a problem?

[SeeMoreDigital]

Originally posted by billou2k
I'm encoding some shorts sequences from TV, I'm storing them as YUV uncompressed AVIs. They are PAL (720x576) pure interlaced and they appear properly in VDUB. Wow... these files must be massive... As a matter of interest, how many MB per min are they?


Cheers

[stephanV]

Originally posted by SeeMoreDigital
Wow... these files must be massive... As a matter of interest, how many MB per min are they?
Cheers

By simple calculation

720*576*25*3*60 = 1780 MB/min for 4:4:4 YUV, so 2/3 of that for YUY2 and and 1/2 for YV12 :)

@Billou2k:

what happens with 2-pass?

[billou2k]

Way to big! :D Roughly 1.15GB a minute (in YUY2) that's why I only keep a few minutes of those ;) I have 2 clips of 5 minutes each and I test them with new codecs (or new versions) in 1 pass mode at several bitrates to get my own idea of how they perform. Full res interlaced is quite hard on codecs ! ;)

[billou2k]

Ok forum is back now.
I've tried in 2 pass mode and it gives the same results:
interlaced encoding in 2 pass in balance mode is fine BUT 2 pass in "better quality" or "extreme" mode cause faulty playback.

[AsTimeGoesBy]

Where is the post-processing configration interface for DivX Fusion please?

Why does the DivX log (map.log) file for a 2-pass encoding have an entry 'nframes=#' whereof this # is equal to the number of frames plus 2?
(It seems that there also are 2 more rows than really existing frames.)

Will there be any tool for DivX Fusion similar to 'EKG' to set personal bitrate preferences?

[LordRPI]

Those dont' exist in consumer space.. yet...

[jggimi]

An updated version of DivX Network's Fusion codec has been announced in http://forum.doom9.org/showthread.php?s=&threadid=88988

To avoid confusion between these two releases (both named Fusion), this thread is being closed. Please post all followups regarding the current Beta to the new thread.