On2 VP3

[stankus]

I just tried the On2 VP3 codec, and the resulting video was excellent! I am very impressed with this codec. It also has been ported to all major platforms, if I remember correctly. Has anyone else tried encoding with it? It even has built-in variable keyframe placement. All that is missing is a 2-pass feature, but the quality is still great without that.

http://www.vp3.com

[bb]

I tried it, too. It's very slow compared to DivX4, though, not half the speed on my Athlon Thunderbird 1,3GHz.

It seems to be quite good at low bitrates. Not a bad choice if you want to use bitrates around 500kbps.

It's one of the better codecs, but I still prefer DivX4.

bb

[stankus]

What do you mean it's slower? Encoding is definitely slower when you encode without using "Quick Encode" mode. Playback, however, doesn't seem to be any slower to me. It uses more or less the same CPU power on my computer (using a Duron 600).

[amni]

This seems to me the only drawback of VP3, otherwise it seems to be
better than any DIVX in terms of quality per file size.
Maybe check again, CPU usage.

I wonder why there is not enough attention in this forum to VP3,
especially sincd it's source is open (which meens improvement is possible
by people who know to program video codecs and utilities0.

Originally posted by stankus
What do you mean it's slower? Encoding is definitely slower when you encode without using "Quick Encode" mode. Playback, however, doesn't seem to be any slower to me. It uses more or less the same CPU power on my computer (using a Duron 600).

[bb]

Originally posted by stankus
What do you mean it's slower? Encoding is definitely slower when you encode without using "Quick Encode" mode. Playback, however, doesn't seem to be any slower to me. It uses more or less the same CPU power on my computer (using a Duron 600).

My test results of this evening:

Athlon Thunderbird 1,3GHz, Win2000, 128MB RAM, 40GB HDD

Cartoon test clip captured from TV had 781 frames interlaced @ 768x576, no audio.
Filters used (I know, without filters and using a progressive source would give clearer results, but I used the same settings in both encodes):
Donald Craft's smart deinterlace, resize 640x480

The results:
------------

VP3 (standard settings, quick mode, 168 kB/s = 1344 kbps):
1. Time 3min 56sec
2. Filesize 5,21 MB
3. CPU usage in WMP: 70%
4. Video: blurrier, less artefacts

DivX4.11 (1-pass, 1347 kpbs, slowest):
1. Time 2min 6sec
2. Filesize 6,23 MB
3. CPU usage in WMP: 50%
4. Video: sharper, more artefacts

Conclusion:
With VP3 you get smaller filesize than with DivX4.11, but you need twice the encoding time. Image quality seems to be better using VP3, but it's a matter of personal taste if you like the blurrier picture or not. Unfortunately there's no 2-pass available for VP3 to exactly match the desired filesize.

bb

[stankus]

I'll have to run a test again. I don't remember at the moment if DivX4 has a CPU meter like MS-MPEG/DivX3, but if you increase that to the highest, you get more smoothing and less blocks, and of course, more CPU usage. If I remember correctly, DivX3 set to max had the same CPU usage as VP3. But, I'll have to do some testing. I am using a Duron 600 and the CPU usage for the clip I tried maxed at about %80 with VP3. I wonder if it adjusts CPU usage automatically?

I also wonder why VP3 doesn't get more attention, amni. That is why I am posting here. Maybe there are a lot of people using it and I just don't know it yet... I like that VP3.2 has the sourcecode available and it's available on multiple platforms, in addition to having really good quality (my test matched WM8 quality, and had better motion handling in some cases).

I think more people should try and it do some tweaking with it and see what happens!

[amni]

called System tool, very convenient, gives CPU usage any second.

I guess all other WINDOWS OS has similar CPU measuring tool.

Originally posted by stankus
I'll have to run a test again. I don't remember at the moment if DivX4 has a CPU meter like MS-MPEG/DivX3, but if you increase that to the highest, you get more smoothing and less blocks, and of course, more CPU usage. If I remember correctly, DivX3 set to max had the same CPU usage as VP3. But, I'll have to do some testing. I am using a Duron 600 and the CPU usage for the clip I tried maxed at about %80 with VP3. I wonder if it adjusts CPU usage automatically?

I also wonder why VP3 doesn't get more attention, amni. That is why I am posting here. Maybe there are a lot of people using it and I just don't know it yet... I like that VP3.2 has the sourcecode available and it's available on multiple platforms, in addition to having really good quality (my test matched WM8 quality, and had better motion handling in some cases).

I think more people should try and it do some tweaking with it and see what happens!

[LigH]

I currently use VP3 quite often for detailed tests. As far as I could find out, it was derived from Duck's TrueMotion 2. In complex scenes it loses details rather quickly, at scene changes details are added progressively (no bad choice: the brain needs a bit time to recognise a new scene). It is more stable to noise and flickering (e.g. encoding old b/w movies from the World War ages - see my notes in this thread). VBR support for VP3 would be great (maybe the commercial VP4 already supports it?). This would be the next step after speed optimisation:

Example: Transcoding "Return of the Mummy" (more than 2 hours) to 512x224 on a Duron 800 in VDub with video filters

- DivX 4.11: around 4-5 hours
- VP3 quick: around 8 hours
- VP3 intense: over 14 hours

[stankus]

VP already has some kind of "bit-saving" functionality which is supposed to use less bandwidth for low-complexity scenes and save that for high-complexity scenes. This isn't exactly VBR, but allowing you to set the maximum bitrate and then having it adjust is more or less what VBR is. Have you tested WM8? It takes about the same amount of time as VP, and again, the quality is comparable between the two. I prefer VP.

[ProfDrMorph]

Looks like most people don't understand what VBR means. It stands for Variable BitRate and just means that the codec uses different amounts of bits every second and not the same amount for every second like CBR encoding does ( CBR = Constand BitRate ).

Since VP3 has different amounts of bits depending on the scene it is a VBR codec.

... but not 2 pass VBR ...

[ProfDrMorph]

... but not 2 pass VBR ...

Right but we were discussing whether VP3 is a VBR or CBR codec and if it has 2-pass support or not.

[stankus]

Originally posted by ProfDrMorph
Looks like most people don't understand what VBR means. It stands for Variable BitRate and just means that the codec uses different amounts of bits every second and not the same amount for every second like CBR encoding does ( CBR = Constand BitRate ).

Since VP3 has different amounts of bits depending on the scene it is a VBR codec.

Right. The only thing is that users can't set the low end (minimum bitrate) but that isn't really important anyway.

Two-pass could easily be achieved via something like NanDub, if it had support for the codec. Maybe it's time to write in a request...

[ProfDrMorph]

hmmm just another thing about VP3: the VfW decoder produces pictures with only 256 colors( at least it looks like that in VDub ) and the Direct Show filter is way too dark.

My system:
Duron 800 MHz
512Mb Ram
Windows 98 ( upgrading to 2k the next days )
Hercules GeForce2 Ti
VIA AGP driver installed ( from VIA 4-in-1 4.36 )
nVidia Detonator 23.11

[EDIT]
Does anybody else has problems with brightness/colors?
[/EDIT]

[TheWEF]

Originally posted by stankus
Two-pass could easily be achieved via something like NanDub, if it had support for the codec. Maybe it's time to write in a request...

if it's easy, maybe it's time to write some code yourself...

wef.

[Baalthazaar]

When you say that VP3 is slower than Divx4.11 (or 4.12, now) you should remember that VP3 is a true color only codec. You only get the real speed improvements in Divx4 when you go to a YUV12 colorspace. I've found that when RGB data is handled with Divx4 its speed is only slightly faster than that of VP3, so I wouldn't call VP3 'slow'.....it just works differently.

[C0mPr355]

Yes i agree more support should be given to VP3.

/me raises VP3 flag in the air.

perhaps i may be able to help some people that would like answers about VP3...since i actually werk for On2 tech

[Taric25]

Okay, is there any reason to use VP3?

In my opinion: NO!

1. Quality: It's actually worse than DivX.
2. Capability: No 2-pass and nowhere near as configurable.
3. Speed: Even slower than DivX.
4. Fileize: Smaller than DivX, but if you lower DivX to a bitrate where the quality is just as bad, then DivX has a smaller filesize.

[bb]

Originally posted by Taric25
1. Quality: It's actually worse than DivX.
2. Capability: No 2-pass and nowhere near as configurable.
3. Speed: Even slower than DivX.
4. Fileize: Smaller than DivX, but if you lower DivX to a bitrate where the quality is just as bad, then DivX has a smaller filesize. [/B]

1. False. Depends on bitrate / filesize. In certain configs VP3 is better.
2. True.
3. True.
4. False: At lower bitrates VP3 quality is recognisable better than DivX, provided that you have the same filesize.

BTW: I'm talking about DivX4 (didn't use 3.11 for months).

bb

[LigH]

Just can agree to bb: VP3 can be better than DivX - I was very impressed in the cases of e.g. "Die Feuerzangenbowle" and "Highlander" - both quite hard to compress and with a low original quality.

I really miss a predictive 2-pass VBR mode: I agree that VP3 actually shows a VBR behaviour - because it does not use more bitrate than required in easy scenes (it has some kind of maximum quality per frame). But predictive 2-pass VBR would also allow to use more bitrate for more complicated scenes than the average target bitrate which is entered in the ICM dialog.

The VP3 behaviour is more related to the ABR algorithm of LAME: it has a small "window" (which is configurable in VP3) and can vary the next bitrate according to the previously used capacity (during the already encoded frames in the lookup window) and the current complexity; in 2-pass VBR, the window can be as wide as the whole file instead (because the whole file's complexity distribution is known).