QuEnc 0.51 (AVS->M2V/M1V Encoder)

[dragongodz]

Yaz - lmin etc are actually libavcodec settings not mencoder(which uses libavcodec the same as QuEnc does). they are available in source of course and Nic may wish to play around with them to see what yields better results.
i dont see why every single option needs to be available for users to change either. i mean you are asking for people to mess with things they shouldnt, result = people blaming QuEnc for their own mistakes. QuEnc is meant to be relativly easy to use but also allow people to change settings customise the output. its a fine line how many of the options are needed for the 2 to stay true. Nic has added options as requested and deemed reasonable so i wouldnt count the current setup as final just yet.

[Amenophis]

I did some more encodes with QuEnc, the quality was fine and all played perfect on my hardware player. nevertheless i've got some observations and suggestions:

- the first frame of every cell has a bad quality. (Althrough it's hard to notice during playback).
First Frame:http://mitglied.lycos.de/enc0404/b2/hl1.jpg
Second Frame:http://mitglied.lycos.de/enc0404/b2/hl2.jpg

- more info about the bitrate peaks: I've gotten some with the settings HQ/Trellis/VBR/Scene Detection using an average bitrate of 3500 kbit/s. the peak is somewhere in the middle of a cell, the scene has high motion and details. Bitrate is 600 kbit/s above the limit. QuEnc encoded this GOP and some GOPs before and behind only with I-Frames. Without Scene Detection these peaks (at least this one) doesn't happen.

encoded frame:http://mitglied.lycos.de/enc0404/b2/matr3.jpg
bitrate viewer:http://mitglied.lycos.de/enc0404/b2/matr3bv.jpg

- does the libav codec provide an option to turn on the non-linear quantizer scale? imho this could improve quality, cce and tmpeg are using this.

- is QuEnc already optimized somehow for interlaced encoding(interlaced DCT, alternate scan,...), or does it just set the interlaced flag? Additionally, QuEnc seems to set no TFF flag even if the source has one, so i've to set this in restream afterwards.

[lamster]

Hi,

The pause feature is handy; might be nice if there was some visual indication that it was paused (other than waiting to see if the screen's stopped updating). Also, the calculation for time remaining doesn't take into acount the amount of time that the process has been paused.

(Neither one's a big deal; just thought I'd mention them.)

Thanks.

[yaz]

@dragongodz
i'm afraid u misunderstood me. i did not want to push nic in any way. he asked sg & i just answered want i'd found about it so far. that's all.
the terms i used is not that used with ffmpeg that's why i mentioned.
what's too much & what's not is quite subjective. e.g. i like if i have control on all(!) the relevant factors. u don't. that's a point where we differ.
i can't really get what u want from me. should i stay silent & keep what i have just for myself, or what ? imho, i've never ever told even a word aginst quenc, so ... (???) :-((
the bests
y

[dragongodz]

i'm afraid u misunderstood me. i did not want to push nic in any way. he asked sg & i just answered want i'd found about it so far. that's all.
the terms i used is not that used with ffmpeg that's why i mentioned.
no it appears its you who has misunderstood. i dont see where i said you were trying to push Nic or that you shouldnt post what you have observed. i pointed out that the options are libavcodec settings so ARE relevant and Nic has the ability to change them in source. mencoder and ffmpeg are basically just front ends to libavcodec same as QuEnc is.

what's too much & what's not is quite subjective. e.g. i like if i have control on all(!) the relevant factors. u don't. that's a point where we differ.
well i already said its a fine line how to keep a program newbie freindly while adding options for more experienced users. ok you and i may know what every option does but someone else just learning is more likly to stuff things up changing things he/she shouldnt if every option is given. as a programmer i know you should never asume that only people that have been using these types of tools for years are going to be the only ones to do so.
i also said that Nic has added options so may add more later aswell so the more control you wish may be there later.

i can't really get what u want from me. should i stay silent & keep what i have just for myself, or what ? imho, i've never ever told even a word aginst quenc, so ... (???) :-((
huh ? i never said not to post your observations(if anything i pointed out how relevant they are) and i never said you said anything bad against QuEnc either.
so what do i want from you ? not a thing. i have no idea why you felt the need to go on the defensive, thats something for you to workout with yourself. :)

[yaz]

@dragongodz
ok ... i've let the steam out. sorry, if i was too offensive. i just felt u blamed me for sg i'd never done. sorry, again.

ok, some positive feedback. i've been experimenting further (with mencoder :-) i've found there's a hard limit for 'limited encodes'. i mean, when not only the target but the limiting bitrates are also interested.
i tried to vary maxrate & minrate simultaneously so as to keep a maxrate=2500. some results

target max min -> avg./max /q_avg
1500 2500 not set 1499/2493/3.52
1750 2500 not set 1749/2868/3.06
1750 2135 not set 1750/2498/3.11 (some buffer underflows)
2000 2500 not set 2000/2895/2.74 (some buffer underflows)
2000 2000 not set 1998/2638/3.03 (heavy buffer underflows!)
2000 2000 2000 2000/2398/3.78 (some buffer underflows)

from here i've found no way of going on by varying min/maxrate. instead i used lmin/rate_tolerance. but neither this way i was able to maintain target=2250,max=2500.
pls, note also that limiting the bitrate means a kinda degradation in the overall quality as indicated by the average quantizer. for my testclip the apparent 'optimal' target bitrate is somewhere round 1750. going up or down from here gives higher average quantizer. so in this instance the rule of 'increase the bitrate for better quality' is not so self-evident.
for going on i must reexamine the other parameters i 'optimized' before. maybe that impose some unnecessaqry restriction(s) on the bitrate business.
the bests
y

[dragongodz]

ye i have been testing different options aswell. the rate control simply is not working properly. this needs major work in libavcodec but we can all hope it will be fixed eventually.

[SILICON]

@Nic

Two weeks without new releases.
Do you stop the development?
Are the bugs very hard of "kill"?

I don't want stress you, but I'd like get some news of your wonderfull development.

Sorry by my impatient.

Cheers.

[Nic]

Ok, the lack of development can be summed up in "I'm busy"
(edited..)

I'm sure you guys understand ;)

Also, got a big project at work that takes precedent :(

I'm keeping a close eye on the libavcodec developments. Alot of the bugs can't be fixed without a re-write of the ratecontrol. Might realease a version later just to get rid of the "Cannot open video codec bug", but without much other change...

-Nic

[BruceL]

@Nic

Sounds like a good reason for a spring break!

Enjoy! :D

(will be patiently waiting)

[jorel]

:cool:
a really great new from d2s readme that i don't knew:
"xx may 2004 version 1.2.2 build 2 (Unreleased)
- New feature: I have added support for Nic's QuEnc. Now DVD2SVCD can be run with freeware programs only if you want to :) "
why nobody post about this here?(or i don't saw?)
maybe this is one of the big great Nic's project! ;)
hey Nic, you edit the b...d? :p

[DDogg]

I'm keeping a close eye on the libavcodec developments. Alot of the bugs can't be fixed without a re-write of the ratecontrol. Glad you brought that up as I'm not so sure that everybody using QuEnc realizes you are dependent on the libavcodec work and there is not much you can do until those updates happen. It's kinda funny really, I'll bet you may get slightly frustrated while waiting for libavcodec updates like other folks get frustrated waiting for an update of QuEnc :)

[Nic]

No, I just get frustrated that I don't have time to do the libavcodec stuff myself....Nothing I'd want to change is particulary difficult, but the testing of a new ratecontrol system would take a lot of time, and I think that's also what's probably stopping Michael from coding it now...

Michael on the ffmpeg-dev list, has a very dry way about him, it's a good read while waiting for any good news ;)

e.g.
For example yesterday he rejected a patch with the only reason being:
"rejected, its 50% cosmetics"

-Nic

[human75]

i make a dvd.my dvd have 2 track audio..m2v maximum bitrate 9000.my audio files english 5.1 + english 2.0 and 1 english subtitle.What is the qenc 0.51 buffer overflow problem.thanks for idea.

[dragongodz]

human75 - this would be the total max bitrate spiking over the dvd specs. try setting max bitrate in QuEnc to 8000 and see if that restrains it any better.

[human75]

ok thanks for answer,Qenc give over flow error,but i make same movie with CCE 2.5 (max bitrate 9000) dvd maestro compile without over flow problem.
9000(m2v)
448 (5.1 DD audio)
192 (2.0 DD audio)
+
-----
9640 kbps...that's right?
what is the problem in QENC?
i try max bitrate 8000 in QENC...and reported.Thanks for ideas...
---------------------------------------------
i try max bitrate 8000.dvd maestro compile without overflow problem...thanks

[dragongodz]

its a known problem that QuEnc(libavcodec rate control to be exact) can spike past what you set for max bitrate. so if you set 9000 it may say spike to 10000 or such. then you go above dvd max total bitrate which is why maestro would complain.

[mr.johnson]

I just downloaded the ffmpeg source. I know you can encode many codecs. But I only find one file ratecontrol.c. The first time I ever looked at such a complex project...does that mean every VBR is done with that? mpeg1,mpeg2,mpeg4?? :confused:
Just a question: But does bbmpeg have such bitrate problems, too?
Because if it worked there..bbmpeg isn't worked on anymore...perhaps this project can merge with ffmpeg?
:stupid:
Please flame for not having had a closer look at it...:eek:

[dragongodz]

does that mean every VBR is done with that? mpeg1,mpeg2,mpeg4?
i havent looked at the other codecs referencing so dont know but its possible.

Just a question: But does bbmpeg have such bitrate problems, too?
BBMpegs rate control is basically the reference TM5 rate control. i havent tested it for max bitrate but it may be an idea to. :)

[freelock7]

Great freeware Nic!
Works fine for MPEG2 encoding with avisynth scripts!
Settings:
-VBR(2000-min)
-treillis:on
-double pass: on(1)
-detect scene change:on
-gop size:15
-high quality:on(2)

(1) "off" make pixels in action scene.
the 2pass create high bitrate in complex scene-the 1pass don't do that!
(2) for action movie always "on" for best result.
Tmpgenc works good but slower than QuEnc.

[mean]

For the 2pass, it is another feature of libavcodec :)
It does the maxbitrate clipping when it does computation at the end of the 1st pass to distribute the bitrate
But it does not try to enforce maxbitrate in the actual pass2
So if the effective frame size differs from the projected one, overshoot happens and cumulates

There is also a patch that makes things a bit better in the same source tree as for the pulldown patch

It is not yet really good, but better than without it, the overshoot is smaller.

[SILICON]

A few monts ago I see one project, don´t remenber somewere. The procet merge a MPEG codec and the XviD engine.

Can you use the controlRate of XviD for the QuEnc?
Is a stupid idea?

[dragongodz]

since Xvid doesnt respect max bitrate at all there is no point using its rate control.

[EpheMeroN]

I just wanted to say GREAT work on this free alternative MPEG-2 encoder.

I have been using it for a month now and am really pleased w/ the results Nic. Keep up the awesome work.

I've only been using the CBR and 1-Pass settings for my MPEG-2 due to hearing there's issues w/ the VBR going above the max DVD bitrate limit. Nonetheless, CBR @ 5600 bitrate is very pleasing to my eyes w/ my analog captures.

One quick question though... under the "Advanced Options" where the GOP Size is, shouldn't that be 18 for NTSC? Or is 18 the max limit for NTSC because that's what it is within the TMPGEnc templates.

I hope to possibly see Min, Avg, and Max bitrate settings in a future release :D

[dragongodz]

setting the max bitrate well under your real desired max can help reduce the spiking. so if you wanted say a max bitrate of 9000 set max at 8000 and see if it gets over 9000 then. :)

GOP size of 18 is maximum for NTSC just as 15 is maximum for PAL.

[EpheMeroN]

Are there any benefits of setting the GOP to the MAX of 18 for NTSC material?

[jdobbs]

You shouldn't set it to 18. I would recommend no higher than 15. There are circumstances in which CCE makes a decision to add more frames than you've set. If you set it to 18, you could end up with an illegal GOP.

[TSO COBOL]

how can I do for VCD ?

I don't want VBR, but I have it even if I don't make "use vbr"

I say that cause bitrate viewer show me lots of variation and mplex tell me that the bitrate is variable and he dont mux

thanks for all

[EpheMeroN]

Originally posted by jdobbs
You shouldn't set it to 18. I would recommend no higher than 15. There are circumstances in which CCE makes a decision to add more frames than you've set. If you set it to 18, you could end up with an illegal GOP.
And by increasing / decreasing the GOP size... what does this do exactly?

[wmansir]

A GOP is a self contained Group of Pictures that does not reference frames outside the group. There are several ways to encode a still frame in MPEG2 and the most efficient rely on what was in the previous frame (P type) or both the previous and next frame (B type). There is also a way to encode a frame so it depends on neither and can be decoded independently (I type), each GOP has to start with one of these.

By increasing the GOP size you increase the number of P and B frames and reduce the number of I frames in your video. Since P and B type encode more efficiently that means your video should look better.* But hardware decoders can choke if the GOP size is too large. My player doesn't even like a 15 frame GOP (at least it didn't with SVCDs), so I stick to 12.

* Of course, this isn't always true.

[jdobbs]

Originally posted by jdobbs
You shouldn't set it to 18. I would recommend no higher than 15. There are circumstances in which CCE makes a decision to add more frames than you've set. If you set it to 18, you could end up with an illegal GOP. Man... I've been discussing questions on CCE in the DVD-RB forum so much that I didn't even look at the title of the thread and forum I was addressing. The fact that CCE does this is of course unrelated to QuEnc. My apologies for the OT answer.

[dragongodz]

I don't want VBR, but I have it even if I don't make "use vbr"
confirmed and noted.

Since P and B type encode more efficiently that means your video should look better.
in low bitrate encodes you may notice a difference for example but in general you probably wont.

There are circumstances in which CCE makes a decision to add more frames than you've set. If you set it to 18, you could end up with an illegal GOP.
that would be known as a bug in CCE then. :)

[TSO COBOL]

I have tested in command line too, same result

you can do something ? :)

[DannyZ69]

Is this still under development?

I believe QuEnc has the potential to be one of the greatest MPEG encoders out there, if this min/max bitrate limitations can ever get sorted out!

[Shubin]

Looks like a great tool, it's fast and quality looks like
great, but...
I'm using dvdauthor+dvdauthorgui latest versions.
Dvdauthor complains about files made by QuEnc, it says
'too many dropped frames' (???) and adds only part of
the file, for some strange reason.
The same .avs file, encoded with TMPGEnc at the same bitrate
does not cause such problems.
What could be done ?

[dragongodz]

Is this still under development?
no new version for a little while and that means its dead ? no it isnt so yes its still under development. Nic is a very busy person though.

you can do something ?
already done some testing. hmm libavcodec doesnt seem real CBR freindly. will try some more things out and report all finding to Nic so hopefully this should be improved next version or so....maybe. :)

Dvdauthor complains about files made by QuEnc, it says
'too many dropped frames' (???) and adds only part of
the file, for some strange reason.
ok i will try to squeeze looking at this in aswell at some point. can you tell me your settings you are encoding to so i can try to emulate it ?

[EpheMeroN]

I'd like to request an option for a future release.

An option to choose either YUY2 'or' YV12 as the input colorspace.

99% of the video I feed to QuEnc is YUY2 and I try to keep max quality and avoid color conversions at all costs buy QuEnc requires YV12.

It's still a great tool. Just a simple and humble req :)

[dragongodz]

ok more testing done with mpeg1 cbr. hmm definatly not good. i even tested using libavcodec's(ffmpeg) -vcd setting. it couldnt keep cbr even using that. yep it still went up and down all the way and infact undersized the target bitrate.

so until rate control in libavcodec is rewritten this may not be really useful.

[Wilbert]

99% of the video I feed to QuEnc is YUY2 and I try to keep max quality and avoid color conversions at all costs but QuEnc requires YV12.
It's not possible to avoid color conversions. There will always be a conversion to YV12, because the video stream in mpeg2 is stored as YV12.

[Paced]

Hey everyone, I think I've come across some sort of bug when "Scene Detection" is turned on - with "2 Pass Encoding" enabled. Firstly, I've made a short (21 second) clip for whoever's interested to test what I've found; an AVS script is included in the file:

Test Clip (4.65 MB) (http://www.geocities.com/cuong_fu/testclip.zip) - Right click, "Save Target As"

Secondly, here are the settings I used on this particular clip:

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

- MPEG-2
- Bitrate = 2641
- Use VBR = ON
- High Quality = ON
- Use Trellis Quant = OFF
- 2 Pass Encoding = On and OFF (I did two tests)

Advanced Options
- Use KVCD "Notch" Matrix = OFF
- "Extreme & Slow" Settings = OFF
- GOP Size = 12
- Max B-Frames = 2
- Interlaced Encoding = OFF
- DC Precision = 8
- Force Closed GOP = OFF
- Scene Detection = ON
- Auto Max Bitrate = OFF, Max Bitrate = 9000
- 4:3 Aspect Ratio

I did two tests with the same settings for both; however, the only difference was: one test was done with "2 Pass Encoding" enabled, and in the other, it was disabled (i.e. 1 Pass).

Here are the screenshot comparisons between the two clips (the file names are pretty much self-explanatory :D):

Screenshot Comparisons (308 KB) (http://paced.port5.com/Screenshot-Comparisons.zip)

As you can see, Scene Detection + 2 Pass Encoding = extremely blocky; and, Scene Detection - (minus) 2 Pass Encoding = perfect.

In addition, if you turn off Scene Detection, but leave 2 Pass Encoding on, the result is much better, but nowhere near as nice as the Scene Detection - 2 Pass Encoding combination.

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

Finally, here is some information for those who will be running a few tests on this clip:

Open the resulting files in VirtualDubMod/VirtualDub-MPEG-2, then scroll to frame 292 in both clips (1 Pass/2 Pass) and compare them. Hopefully, you'll get the same results as I did :)

PS - Very sorry if my Web host is slow for you.

[dragongodz]

Paced - i got the screenshot file ok but the test clip throws a 404 error (page not found).

[dplaton]

By my experience, a limited one, I think that QuEnc is the best.
It will be helpfull if I can use a frameserver to feed QuEnc.
Now I capture the YUV raw with a cheap TVtuner,
I process the avi with VirtualDub compress Huffyuv,
encode with QuEnc and Beeswet,
and multiplex with TMpeg
to produce my SVCDs.

QuEnc work for me faster then Ulead, CCE, TMpeg, and even ffmpeg.
Also QuEnc is much more powerfull because of his elasticity.
My clips are not standard SVCD but with higher average bitrate.

I wasn't able to work directly with the raw avi, maybe because of ffmpeg.

dplaton@home.ro

[Paced]

Originally posted by dragongodz
Paced - i got the screenshot file ok but the test clip throws a 404 error (page not found).

Oops! Sorry dragongodz, my bad, my original Web host wouldn't allow a filesize that big on its server :( It's fixed now; had to make a new account with Geocities (:(), and had to shorten the clip by a few seconds for it to fit on there, hehe. Sorry again.

[dragongodz]

Paced - cool, its downloading now. cant say when i will get to play with it properly since i am sick as a dog at the moment and some things are happening in real life over the next couple of weeks which will mean i have very little free time. since i have it though i will look at it though. :)

[Paced]

Originally posted by dragongodz
Paced - cool, its downloading now. cant say when i will get to play with it properly since i am sick as a dog at the moment and some things are happening in real life over the next couple of weeks which will mean i have very little free time. since i have it though i will look at it though. :)

Cool, no problems. Thanks for taking interest, and take it easy mate :)

[phaeton]

Is there still a problem when using assumefps(23.976)? I'm trying DIKO to backup an XviD to DVDR I have, and I can _never_ get the subtitles to sync when I do it so I'm hoping this can... but when I try to run it when Quenc comes up, it gives me this error (in AVSReader):

Failed on AVS Load!

Does file c:\movie0.avs exist? (Errorcode: -21894..(big number))

Here is my AVS script:


Avisource("D:\cd1.avi",false)++Avisource("D:\cd2.avi",false)
AssumeFPS(23.976)
WIDESCREEN=0
Overscan=0
source_widescreen=0
WIDTH=704
HEIGHT=480
converttoyv12()
BlindPP(cpu=4)
Blockbuster(method="noise",detail_min=1,detail_max=3,variance=0.1,seed=1)
Convolution3D(1, 6, 12, 6, 8, 2.8, 0)
DivXResize(WIDTH, HEIGHT, 0, "LanczosResize",WIDESCREEN)
Undot()
TemporalSoften(2,7,7,3,2)
DCTFilter(1,1,1,1,1,1,0.5,0)
#Blockbuster(method="noise",detail_min=1,detail_max=10,variance=0.3,seed=5623)
DivXBorders(HEIGHT,OVERSCAN)
AddAudio()
MonitorFilter
Converttoyv12()


Can anyone help possibly?

[vmesquita]

phaeton,

Maybe you should post this question in DIKO forum. Anyway, looks like you don't have avisynth 2.54 or the appropriated codec for this movie.

[smok3]

shouldnt there be an option like: 'lower/upper field first' when interlaced encoding is selected? (or is the encoder smart enough to figure out the correct order?)

p.s. iam getting lots of macroblocks on half of the frames on my test clip, fast zoom-outs will break the stream completely..., i guess i must be doing something really wrong.

edit: mkay, vbr must be ticked, no major dropouts now, will test this further. :D
edit2: btw, in the gui - would it be possible to add some sort of label for the pass number where 2pass encoding is selected? (1st pass, 2nd pass)

[Shubin]

Originally posted by dragongodz
ok i will try to squeeze looking at this in aswell at some point. can you tell me your settings you are encoding to so i can try to emulate it ?

- MPEG-2
- Bitrate = 7000
- Use VBR = OFF
- High Quality = ON
- Use Trellis Quant = OFF
- 2 Pass Encoding = OFF

Advanced Options
- Use KVCD "Notch" Matrix = OFF
- "Extreme & Slow" Settings = OFF
- GOP Size = 15
- Max B-Frames = 2
- Interlaced Encoding = ON
- DC Precision = 10
- Force Closed GOP = OFF
- Scene Detection = OFF
- Auto Max Bitrate = ON
- 4:3 Aspect Ratio


Clip contained few seconds of completely black frames.
It's the place where DVDauthor refuses to continue and
where resulting VOB stops playing. Here's what DVDauthor says
about it :

*multiplexing title: 2.....
INFO: [???] mplex version 2.2.2 ($Date: 2003/05/13 20:27:15 $)
INFO: [???] File H:\test09.dvd.m2v looks like an MPEG Video stream.
INFO: [???] File H:\test09.mp2 looks like an MPEG Audio stream.
INFO: [???] Video stream 0: profile 8 selected - ignoring non-standard options!
INFO: [???] Found 1 audio streams and 1 video streams
INFO: [???] Selecting dvdauthor DVD output profile
INFO: [???] Multiplexing video program stream!
INFO: [???] Scanning for header info: Video stream e0 (H:\test09.dvd.m2v)
INFO: [???] VIDEO STREAM: e0
INFO: [???] Frame width : 720
INFO: [???] Frame height : 576
INFO: [???] Aspect ratio : 16:9 display
INFO: [???] Picture rate : 25.000 frames/sec
INFO: [???] Bit rate : variable
INFO: [???] Vbv buffer size : 194560 bytes
INFO: [???] CSPF : 0
INFO: [???] Scanning for header info: Audio stream c0 (H:\test09.mp2)
INFO: [???] MPEG AUDIO STREAM: c0
INFO: [???] Audio version : 1.0
INFO: [???] Layer : 2
INFO: [???] CRC checksums : no
INFO: [???] Bit rate : 24576 bytes/sec (192 kbit/sec)
INFO: [???] Frequency : 48000 Hz
INFO: [???] Mode : 0 stereo
INFO: [???] Mode extension : 0
INFO: [???] Copyright bit : 0 no copyright
INFO: [???] Original/Copy : 0 copy
INFO: [???] Emphasis : 0 none
INFO: [???] SYSTEMS/PROGRAM stream:
INFO: [???] rough-guess multiplexed stream data rate : 0200800
INFO: [???] target data-rate specified : 10080000
INFO: [???] Setting specified specified data rate: 10080000
INFO: [???] Run-in Sectors = 89 Video delay = 13019 Audio delay = 0
INFO: [???] New sequence commences...
INFO: [???] Video e0: buf= 237568 frame=000000 sector=00000000
INFO: [???] Audio c0: buf= 4096 frame=000000 sector=00000000
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=41837 required(DTS)=0
++ WARN: [???] Video e0: buf= 181410 frame=000008 sector=00000224
++ WARN: [???] Audio c0: buf= 2931 frame=000014 sector=00000004
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=85869 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000012 sector=00000522
++ WARN: [???] Audio c0: buf= 4096 frame=000021 sector=00000006
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1276635 required(DTS)=0
++ WARN: [???] Video e0: buf= 181511 frame=000351 sector=00003573
++ WARN: [???] Audio c0: buf= 3143 frame=000585 sector=00000167
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1320521 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000356 sector=00003871
++ WARN: [???] Audio c0: buf= 4096 frame=000592 sector=00000169
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1364553 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000361 sector=00004168
++ WARN: [???] Audio c0: buf= 4096 frame=000603 sector=00000172
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1408438 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000364 sector=00004467
++ WARN: [???] Audio c0: buf= 4096 frame=000606 sector=00000173
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1452324 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000368 sector=00004765
++ WARN: [???] Audio c0: buf= 4096 frame=000613 sector=00000175
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1496356 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000372 sector=00005063
++ WARN: [???] Audio c0: buf= 4096 frame=000620 sector=00000177
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1540388 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000384 sector=00005357
++ WARN: [???] Audio c0: buf= 4096 frame=000641 sector=00000183
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1584566 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000406 sector=00005647
++ WARN: [???] Audio c0: buf= 4096 frame=000676 sector=00000193
++ WARN: [???] Stream e0: data will arrive too late sent(SCR)=1628598 required(DTS)=0
++ WARN: [???] Video e0: buf= 235543 frame=000428 sector=00005936
++ WARN: [???] Audio c0: buf= 4096 frame=000715 sector=00000204
**ERROR: [???] Too many frame drops -exiting

[dragongodz]

sorry i am away for another week yet. just had time to quickly pop in.

shubin - can you try turning off auto max bitrate and set the max yourself and see if that makes any difference ?
also you could try forcing closed GOPs and see how that does.