Latest Nic build, 2 pass encode broken?

[Gaia]

Is it just me or is 2-pass encode broken in latest Nic build? It works ok if you don't use b-frames. Like example if i use 2 and 200% for B-frame settings, 2nd-pass always crashes(you get video compression error) 2 frames before finishing it. If you try default(really stupid thing to do)b-frame settings, 2nd-pass crashes 100 frames before finishing it. I have tried this lots of times with different sources and always same thing happens. Yes i know b-frames don't work currently with QPel or Me hints. Tried it also with lots of different AVS scripts and it still happens.

[Teegedeck]

Strangely enough, I experienced that, too. Only it crashed at once, in one-pass quantizer-based. But using 180% instead of 200% worked and using 8 B-frames worked also, BTW. (I don't remember whether it worked with 8 B-frames at 200% because I just ran a lot of tests and was expecting bugs, so I hardly cared about one or two crashes.)

[Gaia]

Just tried Koepi's build and same.

Second pass crashes 2 frames(2 and 200% b-frames) before end.

DebugView displays this message:
[Virtuald] 2ndpass quant: couldn't read from stats1

So far 1-pass modes have worked in my test. Have to try 8 and 180%.

Update:
Tried 8 and 180% and it crashed 8 frames before end. Same Dbgview message like alway...

Who cares, i have wasted enough time with this. Time to do somenthing else...

[Koepi]

Originally posted by Gaia
Just tried Koepi's build and same.

Second pass crashes 2 frames(2 and 200% b-frames) before end.

DebugView displays this message:
[Virtuald] 2ndpass quant: couldn't read from stats1

So far 1-pass modes have worked in my test. Have to try 8 and 180%.

Update:
Tried 8 and 180% and it crashed 8 frames before end. Same Dbgview message like alway...

Who cares, i have wasted enough time with this. Time to do somenthing else...

Gaia, you're using an experimental codec. If you don't want to "waste time" use a stable build. DON'T BASH ON THE DEVELOPERS OR THE CODEC IF YOU DO THINGS WRONG!

Your stats file is messed up as the error message shows.

Your first pass has less frames then your second pass.

Btw., if you'd checked the resulting file _instead of just bashing on people that can't do anything about YOUR FAULTS_ you'd notice that it plays just fine.

Annoyed by so much ignorance,
Koepi

[Teegedeck]

Gaia, have you checked the location of your .stats-file? In case you're working with several partitions, make sure you're giving an absolute path to the .stats, like "C:\mymovie.stats.".

BTW, I just found the problematic encode with _8_ B-frames at 200% and it is VERY small but still a valid file. It now seems the problem was somewhere in feeding the directshowsource through ffdshow into VDub. Have been jumping onto the bandwaggon too soon...:p

[Gaia]

Sorry Koepi, i am not bashing anybody. I have to admit it was stupid thing to say "waste time" I know it's experimental. Just tried it so many times and same thing always. 1-pass modes works fine, no problems. My stats file is not messed up. I just loaded it to your StatsReader and it's ok. No frames missing.

And believe me i am not doing anything wrong, i have tried several sources and this same thing happens always. I have done about 100 encodes with Xvid, i am not begginer. Have been testing and using Xvid since very first version.I just don't post a lot here... Especially you seem to think if somebody post somenthing negative about Xvid, it's bashing. And believe me this tread was not to meant be negative. This was ment to be a bug report.

My 1-pass don't have less frames than 2-pass, it's just impossible.

[Koepi]

Sorry Gaia for replying that harsh, but you have to understand that I get a little "frustrated" just reading bugreports about development binaries which are officially declared unstable, the "wasted time" made my day there ;)

I'm trying another bframe encode again (my last one 2 days ago worked out just fine) to see if I can reproduce your results somehow.

But i've to see that error myself first, and the only thing about 2pass bframes which is really annoying is that the overflow treatment/curve compression is going nuts (but still hitting desired filesize).

Regards,
Koepi

[Gaia]

No problem Koepi :) I just can't understand why but it happens always. Tried again(done it zillion times allready) for music video and used 2 %200 setting for b-frames and again 2nd-pass crashed just 2 frames before end. This just drives me crazy :) If it helps i am using 98se. If i disable b-frames no problems... I try it again b-frames disabled.

Update:
Same build, same source, same settings, b-frames disabled and no crash :) Funny :)

[Teegedeck]

Gaia, you haven't written anything about the .avi produced in the 2nd pass. Is it usable or just garbage?

[Gaia]

Avi is fine, just frames missing depends what b-frame setting you use.

[sierrafoxtrot]

@koepi

just wanted to say i noticed the same thing but figured ... alpha build. :D

chill out, you're doing a great job, and that stress of bugtracking and development would have driven me into the arms of all that weiss bier by now ...

BTW, qpel looks lovely ... once again, big hugs to all the developers for all these toys to play with :D

sF

[Didée]

With Nic's build of 10-10-02, I didn't experience any crashes up to now. However, I tried B-frames only with 200% and 150%, and only 2 or 8 as maxBframes.

Just to notice: FINALLY I managed to get 'Contact' in really superb quality on 1 CD, with 656*272 resolution (b/p*f = 0.15)!
Bframes do a great job here. Without them, I did 512*208 (b/p*f=0.25), and the result wasn't pleasuring to me...

The magic of XviD.

[Koepi]

I did Contact in 640x272 with stable EPZS - got it looking quite pleasing with the HVS_GOOD matrix of ReferenceDivX :)

Regards,
Koepi

[iago]

@Didée

What B-frames settings did you use for your Contact encode? %200 and 2 max B-frames?

@everyone,

Also, do you (should we) enable "Packed bitstream" and "DX50 B-VOP compatibility" when using B-frames?

thanks,
iago


edit: preparing for a fresh 2-pass encode with B-frames! ;)

[Koepi]

@iago:

DX50BVOPs are a _good_ idea while packed bitstream is not, it drives curve treatment even more crazy :)

Hell, I need a bframe flag in the stats so we can finally do some pleasing 2pass encodes with them ;)

Regards,
Koepi

[iago]

Hell, I need a bframe flag in the stats so we can finally do some pleasing 2pass encodes with them ;) (Koepi)

@Koepi,

Thanks for the reply, friend. And I guess what you mention above will be the next great improvement that will please all XviD users even more! ;)

best regards,
iago

P.S.: Btw, threads about B-frames, QPel, and new dev. builds seem to have grown rapidly and mixed up a lot, which gives me some real trouble to decide on which thread to post to! :)

[Didée]

@ Koepi
I did a few tests (though not too much) with HVS_GOOD, but the results were not exactly my liking. HVS_BETTER is more my taste, but the improvement is relatively little then.


@ iago
Originally posted by iago
What B-frames settings did you use for your Contact encode? %200 and 2 max B-frames?
Yes, Bframes were 2/200, not packed, no div5 compatibilty.

Additional:
unfilter(-100,-100) # Fighting the HEAVVY
sharpen2(0.5) # and ugly Edge Enhancement
BicubicResize(656,272,-.24,.62) # found on the source
# Would be better used on MASK-ed edges only, but I am in a hurry along with
credits @ q20 + bilinear + TempSmooth(4,4) + coring at 64 + grey
intro (space scene) @ 80% bitrate + bilinear
Vorbis @ q2

Question:
Is it really so, that when using Bframes, h.263 quant type is used regardless what the user specifies?