View Full Version : ffvfw
NiTroGen
21st January 2003, 16:01
Originally posted by athos
I have emailed it to NiTroGen too, so he can put it up on his site.
Got it and uploaded it!
Koke
21st January 2003, 19:49
Do I have to remove those two dll`s from
previous build since 2nd pass in VDub
gives me an error? Previous version
works fine.
New version gives normal output in Vidomi
(one pass), old one crashes it.
Thanks
P.S. scarred to remove dll`s from system...:scared:
athos
21st January 2003, 20:05
Originally posted by Koke
Do I have to remove those two dll`s from
previous build since 2nd pass in VDub
gives me an error? Previous version
works fine.
New version gives normal output in Vidomi
(one pass), old one crashes it.
Thanks
P.S. scarred to remove dll`s from system...:scared:
There is no need to remove those dlls, allthough it should not cause any problems to do so assuming you remove the correct ones. I suggest you keep them, as you might install software in the future which needs them. They are for programs compiled with Visual Studio C++ 7.0.
Suikun
21st January 2003, 20:07
This version of ffvfw seems to ignore quantizer restrictions and the b-frame ratio and offset.
On another note, does masking actually do something? Just wondering, as all frames have "clean" avarage quants (like 4.00 and 2.00), which does not happen with XviD's luminance masking.
molerus
21st January 2003, 22:56
Does a msking work at all!!!
I encoded my short clip with the quality set to 100% and using all kinds of masking and it generated THE EXACT filesize as without any boosters.
Do I do something wrong?
athos
22nd January 2003, 00:23
I'm sorry, I dont really know anything more than you guys. I just check out the files with cvs and compile them. I think the best place to find information on this is to study the ffmpeg and libavcodec documentation.
easyfab
22nd January 2003, 22:05
Thanks athos for the build. Very promissing :)
I have just a little question about compiling ffvfw. i try with mingw and i have probs with Tffvfwpage.cpp and tffvfwpage.cpp, do i need directx sdk and something so ?
athos
22nd January 2003, 22:31
Originally posted by easyfab
Thanks athos for the build. Very promissing :)
I have just a little question about compiling ffvfw. i try with mingw and i have probs with Tffvfwpage.cpp and tffvfwpage.cpp, do i need directx sdk and something so ?
I have DirectX 9 SDK and Microsoft Platform SDK (August 2002) installed. For the latest build I also installed Windows Media Format 9 SDK.
I think you at least need DXSDK (8.1 should work too).
easyfab
22nd January 2003, 22:35
Thanks i'll try
gldblade
22nd January 2003, 22:36
Looks very impressive.
My god, anyone care to explain the motion estimate settings? I have an idea how it all works, but what are the specific differences between different settings. For example, X1 vs EPZS for motion estimation.
Suikun
22nd January 2003, 23:05
Originally posted by gldblade
Looks very impressive.
My god, anyone care to explain the motion estimate settings? I have an idea how it all works, but what are the specific differences between different settings. For example, X1 vs EPZS for motion estimation.
Well, for the ME algorithm you can easily test it yourself. Just encode something at const quant 2 (or just do a first pass) and compare the filesizes. The smaller one should in most cases have the better ME, for the ME had found better matches for macroblocks, thus reducing the filesize and improving quality. XviD uses EPZS(²), which seems to be a lot better than X1.
All that SAD, SATD and SSE stuff describe the way the macroblocks of the current picture are compare against the reference frame. XviD uses SAD for example. SATD gives small improvements as far as I know, but is slower. A recent comparison (it's linked somewhere in this thread, I think) showed that DCT gave the best results PSNR wise, but it's even slower than SATD.
Normally this comparison only takes the luminance component of the macroblocks into account, so enabling the Chroma checkboxes will make it do checks agains chrominance, too, which can also give quality and filesize improvements (I guess).
Qpel is, of course, quarter pel motion search precision. It can also improve quality in exchange for encoding and decoding speed.
The high quality checkbox makes ffmpeg encode all macroblocks in all possible modes and choose the best in filesize/quality terms. 4MV allows it to occasionally split a 16*16 macroblock into 4 8*8 blocks for motion vectors, which also can improve quality sometimes.
I can't tell you anything about ME prepass and last predictor count, sorry. But the mentioned test suggested a value of 5 for the latter.
Well, I hope this helps. Someone correct my mistakes, for I certainly have made some...
AlphaDivxMovies
23rd January 2003, 02:34
FFvfw is evolving very fast and they keep on adding more funcionality!
Seems colossal work compared to other projects.
With the latest builds some facts are present:
The best psnr results seem to derive from these settings:
- Trellis Quantization is quite efficient (alone or otherwise)
- From ME side EPZS is the only viable and stable option
- SATD is efficient and relatively CPU friendly but for the very last bit use DCT, slow but with slightly better results.
- Always use 4MV and HQ specially with B-frames.
- B-frames: Use no more than one and avoid them with animated content.
Seems a little too quirky to me so i avoid using them ( 2CDs :))
- Qpel seems promissing but is broken at the moment. Color bleeding everywhere. Next build maybe?
- ME prepass and last predictor count: I use always but after every I-frame works fine and seems less cpu intensive, 3 to 5 on last predictor count works ok for me.
- Works fine in YV12.
- "EPZS diamond size doesn't seems to provide an noticeable gain."
- "Never use v4mv alone, not only you will loose speed but also PSNR!"
- "vhq is only marginaly useful for most scenes, but on the anime one (PM) it's a really huge win!"
-"one B frame + SATD (*cmp=2) wins, sometimes by a large margin. "
For more detailed info check out: http://rguyom.chez.tiscali.fr/libavcodec_tests.html
Seems quite promissing, i did a movie test and it gave better quality thant Xvid at Quantizer 2 and all 1 pass modes, i just can`t get 2-pass to work on my computer, pity, i do first pass ok but second pass crashes all over the place.
Regards
AlphaDivxMovies
sam_b
23rd January 2003, 02:55
I found no colour bleeding after *unticking* Autodetect Bugs in latest ffdshow.
AlphaDivxMovies
23rd January 2003, 03:18
But there is still a bug and likely to be corrected. :)
gldblade
23rd January 2003, 06:47
Thanks suikun.
What does last predictor count do?
athos
28th January 2003, 22:58
New build, changelog:
2003-01-28 14:03 milan_cutka
updated libavcodec
2003-01-28 13:50 milan_cutka
working on presets 3
2003-01-28 07:39 milan_cutka
no message
2003-01-28 06:50 milan_cutka
working on presets 2
2003-01-27 15:37 milan_cutka
working on presets
2003-01-27 06:47 milan_cutka
few fixes
2003-01-24 15:47 milan_cutka
no message
2003-01-23 16:12 milan_cutka
decoding support for HuffYV12
2003-01-23 15:30 milan_cutka
theora cleanup, mpeg1 muxer
2003-01-23 07:06 milan_cutka
huffYV12 encoding, graph currently disabled
2003-01-22 17:17 milan_cutka
saving ffdshow settings to bytestream (no loading yet)
2003-01-22 16:13 milan_cutka
wmv9 complexity settings, alligned allocators for lavc
2003-01-22 06:44 milan_cutka
no message
2003-01-21 15:08 milan_cutka
trellis quant not allowed for mjpeg
2003-01-21 06:39 milan_cutka
reading properties
Should be up at http://athos.web1000.com/ffvfw.html
Direct Link: http://athos.web1000.com/ffvfw-20030128.exe
CavalloPazzo
29th January 2003, 21:28
As usual the link doesn't work... at least for me. I'm waiting for Nitrogen to put a link into his site.
NiTroGen
30th January 2003, 18:18
Originally posted by CavalloPazzo
As usual the link doesn't work... at least for me. I'm waiting for Nitrogen to put a link into his site. Sorry, but I can't get this build from Athos' page to put it on my site. Athos, can you email it to me?
AlphaDivxMovies
30th January 2003, 19:42
I have sucessfully (methinks at least) used the latest version of Ffvfw in 2-pass mode and used Koepi`s Xvid StatsReader to scale the resulting stats file and adding keyframes to chapter locations.
I used no b-frames, no Qpel and no masking, i used a trailer (Intimacy) and a full movie (Final Destination) and in both i hit the desired size right on! I only had a 124kb variation in the trailer and 340kb for the full movie (in excess). Visual quality surpasses IMHO Xvid, Divx and Divx311, it looks very very good!
Doing another 2-pass mode on Final Destination but with B-frames and Masking, i do not like the Qpel feature, seems too quirky to be used in high quality encodes. Will post results later perhaps a bit more detailed.
Regards
AlphaDivxMovies
PS - @athos - I do not mean to be annoying but could you say how you compiled Ffvfw? I can checkout the latest CVS but VC7 refuses to compile it (i have nasm installed), a simple pointer would be enough, i just can`t find info elsewhere, sorry and thanks in advance!
PS2 - Thanks for your latest compile!
athos
30th January 2003, 21:54
NiTroGen> Sorry, I forgot. Mailed it now.
AlphaDivxMovies> You need DirectX SDK, and also Windows Media 9 SDK (for WMV9 support). You might also need Microsoft Platform SDK and possibly STLPort (not sure about these).
gldblade
30th January 2003, 22:37
As usual the link doesn't work... at least for me. I'm waiting for Nitrogen to put a link into his site. Are you left-clicking on the link? Because if you right-click save, or if you use a download manager, you'll simply be downloading a webpage that says that direct linking isn't allowed. If you left-click, you'll get this webpage as well, but you'll also get a download link that works.
NiTroGen
30th January 2003, 22:40
Originally posted by gldblade
Are you left-clicking on the link? Because if you right-click save, or if you use a download manager, you'll simply be downloading a webpage that says that direct linking isn't allowed. If you left-click, you'll get this webpage as well, but you'll also get a download link that works.
This is the message I get when I left-click:
We are sorry, but the file you are trying to download is larger than the maximum length permitted under current network conditions. Please try again later when network conditions change. The file you were trying to download was 501413 bytes.
Anyway, you can also get the binary from my site now.
Koke
31st January 2003, 06:00
Hi!
Just informing that this new
build can`t use preprocessing
option from ffdshov (jan 03 2003)
It is greyed. Previous builds could use it.
Regards
unixfs
1st February 2003, 12:24
Hi,
can you add decoding support for mpeg[12] in avi, please?
Thanks.
Tommy Carrot
3rd February 2003, 15:35
Wow! Let me praise this thing!
I'm very impressed with ffvfw's low bitrate (quantizer 6-10) handling.
I've always hated how xvid and divx(4-5) creates texture flowing (sorry, i cannot explain it better) at low bitrates. It's not a bug, just the way they handle motion.
FFVFW remains clean of this. I did a comparation with xvid, and detail-wise they are about the same, but FFVFW is free from this floating/flowing texture thing.
And i like how many options it gives. Speed-wise it's slower, but i prefer quality over speed (within a reasonable limit of course).
At higher bitrate (2cd rip) xvid is the better, because it's faster, and the above mentioned problem disappears, but at low, ffvfw is the winner by a wide margin.
The only problem i've got: every key-frame causes strong blocks, which disappear only several frames later. At fix quantizer, xvid's keyframes are more detailed.
I've used ezps, chroma, sad, qpel, size 1 diamond, h263, but no bframes and trellis quantizer(i've heard it's a very good option, but i found no quality gain with it, just great speed loss).
Billy|B-A
4th February 2003, 17:08
I may have a small bug, or maybe its something I am doing wrong. When I finsih the 1st pass and try to start the 2nd pass it will crash cause the stats file is in use. The codecs is still holding the rights to the stats file. But if I close vdub, then reopen it I can encode using the stats file.
Its a small problem, but its a irritating 1. Keeps me from queuing jobs on the same stats file.
VirtualDubMod/VirtualDub
ffvfw-20030128.exe
Windows XP Pro
[edit]Took out a line. guess i am not asking for help =\
bilu
6th February 2003, 12:17
This codec seems very interesting, I will try it ASAP :)
Anyone found any suggestions on masking?
Bilu
bilu
7th February 2003, 13:55
Well, I haven't tested yet, but I've been a good boy :D
After reading this on the ffmpeg mailing list...
====================================================
From: Felix Buenemann <atmosfear@us...>
Re: libavcodec documentation
2003-01-30 05:29
On Wednesday 29 January 2003 17:43, Michael Niedermayer wrote:
> Hi
>
> On Wednesday 29 January 2003 12:06, Philippe Ney wrote:
> > Hi all,
> >
> > Is there some libavcodec documentation somewhere?
>
> hmm, not much AFAIK, except the source itself, for example see avcodec.h /
> apiexample.c in ffmpeg CVS or vd_ffmpeg.c / ve_lavc.c in mplayer CVS
add DOCS/tech/libavc-options.txt to the list (from mplayer), it's only about
en/decoding parameters though.
>
> Michael
--
Best Regards,
Atmos
===============================================================
... I've found some docs on libavcodec (the base of ffvfw)
encoding-tips.txt (http://www.mplayerhq.hu/cgi-bin/cvsweb.cgi/~checkout~/main/DOCS/tech/encoding-tips.txt?rev=1.2&content-type=text/plain)
libavc-options.txt (http://www.mplayerhq.hu/cgi-bin/cvsweb.cgi/~checkout~/main/DOCS/tech/libavc-options.txt?rev=1.38&content-type=text/plain)
Edit: I removed the attachments and put the links instead.
Bilu
bilu
9th February 2003, 10:54
Originally posted by Billy|B-A
I may have a small bug, or maybe its something I am doing wrong. When I finsih the 1st pass and try to start the 2nd pass it will crash cause the stats file is in use. The codecs is still holding the rights to the stats file. But if I close vdub, then reopen it I can encode using the stats file.
Its a small problem, but its a irritating 1. Keeps me from queuing jobs on the same stats file.
VirtualDubMod/VirtualDub
ffvfw-20030128.exe
Windows XP Pro
[edit]Took out a line. guess i am not asking for help =\
I have the same problem too, even using AVS2AVI. Same version and same OS.
Bilu
NiTroGen
9th February 2003, 15:38
@Milan, Athos:
Although the latest libavcodec release is version 0.4.6, in the change log (http://ffmpeg.sourceforge.net/changelog.html) you can see that version 0.4.7 has been released, but not mentioned anywhere:
version 0.4.7:
- qpel encoding (Michael)
- 4mv+b frames encoding finally fixed (Michael)
- chroma ME (Michael)
- 5 comparission functions for ME (Michael)
- b frame encoding speedup (Michael)
- wmv2 codec (unfinished)
- user specified diamond size for EPZS (Michael)
I think that currect CVS snapshot (http://ffmpeg.sourceforge.net/cvs/) is version 4.0.7. Is it possible to have a new build using this version of libavcodec?
easyfab
9th February 2003, 17:58
This "version" (0.47 change every day) is already in ffvfw.
Milian make an update very often and as you can see the latest ffvfw has b-frames, chroma me ...
NiTroGen
9th February 2003, 19:30
Check the latest build's about box:codec build date: Jan 28 2003, 22:40:54
libavcodec_enc: version 0.4.6, build 4654 (Jan 28 2003, 22:09:03)
XviD: API version 2.1
easyfab
9th February 2003, 19:56
Originally posted by NiTroGen
Check the latest build's about box:
Yes it's indicate 0.4.6 but it's 0.4.7
Just look the file "version" in the ffmpeg cvs directory it's already 0.4.6
I think they forgot to change it.
Milian updated libavcodec for 6 days now (look in the ffvfw cvs directory)
athos
9th February 2003, 21:38
Well I put up a new build today, I'm not sure if libavcodec is 0.46 or 0.47, about box says 0.46. also mailed it to nitrogen
NiTroGen
10th February 2003, 03:11
Thanks Athos for the new build, but I had no mail from you yet! :( And as usual, I can't access you page. :angry: Is it possible to mail it again? :D
Thanks again!
BTW, how about changing the provider of your web space, since may people cannot access your page. ;)
Sgt_Strider
10th February 2003, 04:42
what does this program do? I went to the ffvfw's official page and still can't find any info to help this noob understand what it does.
athos
10th February 2003, 08:42
i mailed it again. i will look into putting it on another webspace when i get home from work.
NiTroGen
10th February 2003, 11:25
OK, I got it now. I'll put it on my page tonight.
Tommy Carrot
11th February 2003, 02:02
Originally posted by Sgt_Strider
what does this program do? I went to the ffvfw's official page and still can't find any info to help this noob understand what it does.
It's an mpeg4 codec, just like xvid/divx. With much more options, so the chance of confusion is large, but some of that option is amazing (switching off ME, for example). :D
EDIT: ME=Motion Estimation, not 'me'. :D
Sgt_Strider
11th February 2003, 02:47
Originally posted by Tommy Carrot
It's an mpeg4 codec, just like xvid/divx. With much more options, so the chance of confusion is large, but some of that option is amazing (switching off ME, for example). :D
EDIT: ME=Motion Estimation, not 'me'. :D
Is it actually better than xvid? I was getting the impression that ffvfw is some sort of program that compliments xvid.
sam_b
11th February 2003, 03:55
Try it - preferably at several quality settings and compare. Most of the settings you need to know about are fairly self explanatory. It (in use) has nothing to do with XviD AFAIF. Are you thinking of ffdshow?
You might want to pass on b-frames at least at first.
digitize
12th February 2003, 03:10
Suikun was nice enough to tell me a few things about it, and really it's an alternative to divx 3.11. It's better than divx3 and outputs compatible bitstreams. I trust Suikun is right about it, he's a good encoder.
Suikun
12th February 2003, 18:14
Originally posted by digitize
Suikun was nice enough to tell me a few things about it, and really it's an alternative to divx 3.11. It's better than divx3 and outputs compatible bitstreams.
There's actually one thing about encoding to DivX3 with ffvfw you should know about: DivX5's DivX3 playback is very, very broken (at least it was with 5.02, dunno about 5.03). There might be strange blocking artifacts on scenechanges and high motion scenes.
I didn't notice that on my first encodes, so imagine my rage when I noticed that DivX5 screwed up all my work.
A workaround for this is setting the FourCC of the avi to MP43 (remember: DivX3.11 is nothing but a hacked M$-MPEG4v3 codec. The internet's most used codec was made by M$, whee!). _Almost_ everyone should have M$'s MPEG4 codecs installed.
Yet, ffvfw is THE encoder if you're bound to DivX3 compatibility. It's ME is waaay better, it doesn't introduce that annoying smearing (DivX3's block skip decision SUCKS!) and it uses XviD's 2pass algorithm.
As for ffvfw's relation to XviD: In the "Encoder" option you can choose to either encode with libavcodec (see www.ffmpeg.org) OR XviD, in which case ffvfw will use xvid.dll. ffvfw is just a frontend for these encoders.
digitize
12th February 2003, 20:58
@Suikun
Yeah i remember that happened with asagiri no miko for you, and i noticed stony used mpeg 4v3 for hng.
Tommy Carrot
13th February 2003, 01:23
Originally posted by Sgt_Strider
Is it actually better than xvid?
I've wrote about my opinion earlier it this topic in a greater detail. I find it similar or slightly worse at high bitrates, at low bitrates (quantizer 5-10), ffvfw is much better imo. Much less artifacts. Of course the different options can greatly alter the result.
athos
13th February 2003, 17:03
Note: I have had some trouble with gabest's excellent Media Player Classic, and after some advice posted in the MPC thread I uninstalled ffvfw and now it runs fine. I'm not sure if this is only related to the latest builds of ffvfw, as gabest himself says he has xvid, ffdshow and ffvfw installed, and no problem with MPC.
Also, for some other guy uninstalling ffvfw did not help.
bill_baroud
13th February 2003, 17:51
yeah, i had some problem with MPC, i didn't why (mpc launch, use 100% CPU but never show itself). I have just deinstalled ffvfw after read athos comment and it's fine now.
btw, i couldn't achieve a 2nd pass int with ffvfw (20030209) both with vdubmod & vdub 1.4.13 and avisynth 2.07/2.5beta.
First pass were ok, but 2nd crash with an "maybe corrupt video data (-100?)" :(
i was eager to test some new build after tests some months ago (with the first public build by milan).
btw, i'm running XP home(damn) on a compaq laptop 1800+ 256DDR.
no directx 9, no sp1.
Kurosu
13th February 2003, 18:26
I have seen that, when MPC is launched, an enormous amount of memory (around 550MB under Win2K SP3) is suddenly used (MPC doesn't seem to take that much memory, though).
Same happens with DVD2AVI, but not VirtualDub.
@bill_baroud
Did you try to close Vdub, restart it and restart the 2nd pass ?
I think the codec doesn't close the stats file, and untill the codec instance is unloaded, the file is locked in write-mode (can't be read).
Suikun
13th February 2003, 20:49
Originally posted by bill_baroud
First pass were ok, but 2nd crash with an "maybe corrupt video data (-100?)" :(
In some builds there is a bug which prevents ffvfw from starting encoding twice in one VirtualDub session, so after you abort or complete an encode, you will need to restart it.
OntzA
15th February 2003, 01:29
There's a solution to the MPC issue. When you install ffvfw don't tick Directshow filter. It seems MPC doesn't like ffvfw as directshow filter :(
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.