The devs have finally decided it is time to release an XviD 1.0 beta to get the masses testing it to iron out any final bugs that might remain.

Get it from here.

Post bug reports here.

Update: You can now get beta 2 from the same place. Post bug reports here.

Thanks

They warn that the bitstream may still not be MPEG-4 compliant, so better not do any archiving with this yet, until it's more tested.

And let the testing begin! There is sooo much stuff to play with in this build. If you wanted you could enable GMC, qpel, bframes,chroma motion, mpeg quantization, trellis quantization, vhq4, adaptive quantization. So, yes there are a lot of things to play around with. The only thing you should not use is packed bitstream since it is still broken, everything else should work fine. Here are settings I would recommend though for a encode please note that you can tweak by adding and disabling things but this is what I would start with, but first hit load defaults then try something like this:
H.263 Quantization
B-frames 1, 1.50, 1.00
Quarterpel (only if your cpu can handle it and if hardware compatibility isn't a huge concern)
Chroma Motion
VHQ1 (you can use higher than 1 to get even better quality but I think this is a good compromise between quality and speed)
Trellis Quantization
Max I-frame 10xfps
This should give nice results, you can try using adaptive quantization on low bitrates and custom matrices etc etc. and high quality mpeg matrices on 2cd encodes. In general, global motion compensation won't give a huge increase in quality so I wouldn't enable it. Also, don't use Reduced Resolution encoding because it is only compatible with mpeg4 ARTS profile and not ASP. Cartoon mode can be extremely useful for cartoons of course and anime. What it will do is enable static motion detection so that frames can be skipped more often if they are the same. If used in combination with dup in avisynth it can give very nice results at very low bitrates.

nice to know, that there is some semi final work done on xvid (something like 1.0 sounds generally more encouraging to use than 0.9 dev.)...

@ admins (JohnV?): where should we post messages regarding important new codec builds, video apps etc, in validated news submission, or in digital a/v news?!
some posts where moved in the past, some not... so please, give me (us) a statement for further submission.

A/V news is fine. If it's remarkable release (like imo this is, although it's beta, but still waited very long), it may end up to the portal.

I can't find the B-frames settings. Are they hidden somehow, or am I just blind?
Edit: I am blind.

Norman

While the celebration is going on (and I wish the developers could really throw a party and enjoy the glory of their work) here's just some info I believe that belongs into this news-thread:

1) Posting directly to the XviD-devel list is not such a good idea IMHO; it's a list exclusively for developers as the name indicates. Instead I would encourage you to post bug-reports at forum.doom9.org. There we can talk problems through and also filter out reported bugs that really don't have to do with XviD.

2) GMC doesn't help anything with VHQ (you notice when you compare filesizes).

3) For a very, very long time now, VHQ=4 has been the recommended setting. Only in the very beginning with old VHQ-code it had been recommended to use VHQ=1.

Cheers,

Tee

And again, GMC + VHQ 4 is a good combination. It provides a small increase in quality. If you're aiming for a speed/quality compromise, then maybe you can consider disabling it but it certainly does help. Comparing filesizes doesn't make much of a difference if you're aiming for the same target filesize...

By the way, all of us in the xvid IRC channel use VHQ 1 for the first pass and VHQ 4 for the second pass for a speed boost and no effect on the quality.

My recommended settings (for PC playback, obviously.. )

MPEG Custom: HVS Best
QPel
GMC
B-frames (1/1.50/0.00)
Chroma Motion
Trellis Quantisation
VHQ 1 for the first pass, VHQ 4 for the second pass
Always 2-pass mode.
Occasionally Adaptive Quantisation is useful to save some bits.

Use FFDSHOW for decoding!

That's all there is to it really. :-) Enjoy!

GMC is there to reduce filesize by someting like 2-4%. Which it doesn't if combined with VHQ - those two techinques get in the way of each other and leave an unoptimized motion-search. Always compare features in fixed-quantizer mode, first. There always have been reports about perceived increases in quality with GMC - even with the very first code with only two warp-points that really shouldn't have done anything good. That's why I'm a bit sceptical here. I wonder what a blind-test would have shown there.

The use of both VHQ modes is a good idea to speed things up.

Edit: Lots of editing in this post; sorry that I'm such a slow thinker...

Edit: sysKin has corrected me on the fact that GMC now works with VHQ, how shamefull...

Wow.... WOW

Well let the testing begin, any XviD 1.0 Beta's to DivX 5.1.1 comparison's yet (kidding).

Really this is great that the XviD team have got the 1.0 release almost ready.

I assume this is your recommended settings for 2-CD or more encodes. For 1-CD encodes HVS Best matrix and Qpel aren't necessarely the best options I think..

Is interlaced mode working for anyone? I get completely screwed output, when I activate that option (colorful macro blocks, etc.).

What is your personal recommendation for HQ in one CD ?

Thank You John.

Nick

Well for 1CD I'd try pretty much the same what already has been said:
H.263 Quantization
B-frames 1, 1.50, 1.00
Chroma Motion
VHQ4
Trellis Quantization
Max I-frame 10xfps (250 with PAL)
Adaptive quantization is worth testing, and 2 consequtive b-frames also, if the movie is relatively long. GMC maybe if speed is not an issue.

Depending on the compatibility needs, I'd choose between unrestricted, Advanced Simple Level 5 (AS@L5), or one of the 2 highest divx profiles, though divx profiles don't allow GMC (or qpel).

Aiyaiyai! Reminds me of the r3mix/command-line switches insanity with LAME all over again!

Is there some archive/reference for ideal encoding settings for various intents (ie. 1CD/2CD encodes)?

Well.. I really don't think those settings are anything very controversial:
* h.263 quant type - has always been recommended for low bitrates 1CD (instead of MPEG quant). h263 quant produces slightly softened picture.
* B-frames 1 or 2, pretty much default b-frame settings for 1CD. Additional settings very close to default
* Chroma Motion has been recommended for a long time
* VHQ4 - the slowest and best search mode
* Trellis Quantization - Many people liked this already in Koepi2406 build with 1CD
* Max I-frame - as recommended everywhere

Adaptive quantization = old Luma masking should now be working ok, and first reports are good, so it's worth testing. GMC - global motion compensation may reduce bitrate need a bit, but it's slow.

Anyway, as you can see, I offered pretty much the most often recommended settings you can find by reading XviD guides or doom9. Not any "from the hat" settings.

Thanks for the detailed explanations.

Still, though, the situation reminds me a lot of r3mix. It'd be nice if someone wrote a wrapper for these options that allowed you to set a quality value, be it like MPC's --quality x or some target size, and have the options automatically adjust to give the highest quality.

Ah well... Not quite as critical as audio transparency.

A wrapper isn't being planned but my good friend CruNcher and I intend to do some speed/quality testing sometime to build up some XviD profiles for people to use according to how long they're willing to wait for their video. :-)

H.263 quants are generally recommended for 1 CD and Qpel off as well. But I use my settings because they suit my taste. :-)

JohnV: I have tested my settings on three 2 hour+ films and all they all came out looking good at 700mb size. After audio I had about 630mb left for the video. Whether H.263 quants would look better or not is a different matter and probably one of taste. CruNcher did a brief test on Pearl Harbour at about 500kbps for the video and I preferred the HVS Best version (all used QPel, GMC etc). You can find it over at doom9 and have a look yourself if you like. :-) I thought the HVS Best one looked slightly sharper and 'better' to my eyes than the rest and there weren't many blocks apart from on the faces during the second half of the clips CruNcher uses but this was exhibited by all four quantisation techniques. Thanks for your input though and h.263 is the safe way to go for 1 CD. My method is controversial. :-)

XviD-1.0-Beta2-05122003 released.

See also Doom9 thread.

Changelog:
- Fixed bitstream headers to be mpeg4-compliant again.
- Quant matrices are finally threadsafe.
- Interlaced encoding is fixed.
- VFW frontend fixes and misuses of core.
- Final tweaks to frame decision and motion estimation.
- Fixed overflow degeneration (Q31 problem with i-min-distance >1).
- Fixed CBR mode - the first frames weren't capped as wanted.