Which DV Codec? Need a definitive answer, please!

[midiguy]

Hello. I recently purchased a new Panasonic DV-GS70 miniDV camcorder. I have read in the past about the different DVD codecs that are out there, and there pros and cons, but after coming back to this forum (and after examining a plethora of new posts) I have realized that there is no definitive answer. So I am requesting some guidance, because I really have no idea when it comes to DV codecs. I edit in premiere but don't mind using any programs to capture the footage (as long as the footage will work in premiere without a problem). Thanks!

[SomeJoe]

What is the final destination of the video? i.e. going on DVD, tape, etc.?

If you're not going to make a DV .avi for viewing on the computer, or send the DV back out to DV tape, then the codec you use doesn't matter a whole lot.

For instance, take a look at this path, which is how I do a lot of video:

1. DV camera: It takes the video and compresses it with it's hardware-based DV codec inside the camera, and writes the compressed video to tape. You can't control this, the codec is fixed inside the camera's circuits.

2. The capture is just a file copy operation from the camera's tape to the computer. So you can't do anything here either.

3. Editing in Premiere 5.x and 6.x is all done in uncompressed RGB color space. (Premiere Pro uses uncompressed YUV). Premiere never recompresses video using a DV codec unless you export to a DV .avi or export back to DV tape.

4. I then frameserve from Premiere 6.5 (using the Premiere frameserver from www.videotools.net) to AVISynth. This is an uncompressed RGB handoff.

5. In AVISynth, I use ConvertToYUY2() to convert colorspace to something AVISynth-friendly, and apply various filters that I want like Convolution3D.

6. The .avs files I use directly in CCE Basic to compress to MPEG-2 for DVD.


As you can see, no DV codec is ever used on the PC for compression. Premiere only decompresses during editing and frameserving (using the Microsoft DV codec). The Microsoft DV codec has some problems with compression, but that's not being used. (BTW, Premiere 5.x & 6.x use the Microsoft DV codec, Premiere Pro uses the Main Concept DV codec).

[midiguy]

For now, I just need to be able to import footage into premiere, do my edits, and then send it back to the DV tape. I will be using transitions and what not, so I assume it will have to recompress it in some places. So which codec/method should I use? Thanks for your help, by the way.

[SomeJoe]

Ah, so you want to send it back to DV tape. Yes, any transitions or effects will have to be recompressed with a DV codec.

I've had limited use of the Main Concept DV Codec (http://www.mainconcept.com/codecs.shtml), and it seems to work well. This is also the DV codec that Premiere Pro uses internally.

However, you can't directly "Export to Tape" with Premiere 5.x or 6.x and get Premiere to use the Main Concept codec (or any other 3rd party DV codec). Premiere 5.x and 6.x is hard-coded to use the Microsoft DV codec when doing this operation. So what you have to do is export the edited project using Premiere's standard Export Movie dialog, and specify the Main Concept DV codec as the video compressor. You then have to send the DV .avi file to tape with a different capture/export utility.

There is apparently a hack for Premiere 6.x that will cause it to use a different DV codec instead of Microsoft's, thus eliminating the work around I just mentioned. I've never tried to use it -- it looked complicated to install and there were reliability questions.

[midiguy]

Thanks for your help.

[mb1]

Don't use Mainconcept DV. It is seriously buggy.
Even the first render generation produces ugly horizontal lines (v2.4.3 within Premiere Pro and v2.4.4 full one). This is no Premiere issue.
http://mb1.svcdfan.de/small_comparison_of_dv_codices.htm

My way for PAL DV which is YUV 4:2:0 to PAL DVD mpeg2 which is also YUV 4:2:0

- capturing with Let's Edit (ScenalyzerLive writes structurally wrong Canopus DV) [YUV 4:2:0]
- editing with Let's Edit which completely works within YUV 4:2:0 [YUV 4:2:0]
note that Premiere Pro doesn't work internally within YUV even if Adobe stated that
- rendering a new dv.avi (1. render generation) from Let's Edit timeline [YUV 4:2:0]
- record that dv.avi back to dv tape [YUV 4:2:0]
- feed Procoder 1.5 with dv.avi to produce PAL DVD mpeg2 [YUV 4:2:0]

[taudule]

Hello mb1,

ScenalyzerLive writes structurally wrong Canopus DV

Could you please be more precise ?
Thank you

O.

[SomeJoe]

Originally posted by mb1
Don't use Mainconcept DV. It is seriously buggy.
Even the first render generation produces ugly horizontal lines (v2.4.3 within Premiere Pro and v2.4.4 full one). This is no Premiere issue.
http://mb1.svcdfan.de/small_comparison_of_dv_codices.htm

note that Premiere Pro doesn't work internally within YUV even if Adobe stated that

Very interesting comparison of the codecs. Yes, certainly it can be seen that Main Concept has some problems with the horizontal lines.

However, all of your tests were done with 4:2:0 PAL. I wonder if the horizontal lines are no longer visible if working in 4:1:1 NTSC ? Perhaps Main Concept fares better under that condition?

Having output clamped to 16-235 is not a big deal if you're working with broadcast or otherwise camera-derived footage. That footage should already be 16-235 anyway, and is required to be 16-235 for broadcast after processing.

Also, Adobe and Matrox (who just came out with RT.X100 drivers for Premiere Pro) state that Premiere Pro works internally in YUV vice RGB. Now, it's probably not 4:2:0 YUV, the chroma is probably upsampled to full resolution. But it was my understanding that they chose to work in YUV to remove the RGB<->YUV color space conversions. If this is not the case, where/how did you learn of this?

[DSP8000]

You can try MainConcept DV Codec v.2.4.9 if you have the new MC encoder v.1.4.1.I find this codec to be very reliable & fast.
About Canopus Let's Edit, how do you find this program,is it stable,fast,easy? I'm asking because I'm planing to buy it plus Canopus VideoFX.I like Vegas but from what I can see on the Canopus website Let's Edit + VideoFX looks very promising.
Anyway to install MainConcept DV Codec v.2.4.9(from MC encoder 1.4.1) you have to do some registry tweaks as described here on this forum.Works good for me.Can someone do a comparison with the new MC codec?
DSP8000

[theReal]

AFAIK the only difference between Microsoft DV, Canopus DV and Mainconcept DV is the rendering speed. There's no difference in quality because the codecs need to be 100% compatible with what the camcorders use. So, I have always been using MS DV because most of the time you will only render a few percent of the file (transitions, titles), all the rest of the video will only be direct stream copied and the speed solely depends on the speed of your hd(s)

[bb]

Originally posted by theReal
AFAIK the only difference between Microsoft DV, Canopus DV and Mainconcept DV is the rendering speed. There's no difference in quality because the codecs need to be 100% compatible with what the camcorders use.
There are certain differences, mainly bugs with color handling, but a few other issues as well. Canopus uses a different fourCC. There are differences in the DV algorithm implementation and the codec type (VfW vs. DirectShow).

For regular processing all above mentioned codecs are good, and there are ways to circumvent certain shortcomings (for instance correction of 4:1:1 color space interpolation).

bb

[mustardman]

Having used the Miro$soft DV encoder (spit), I must say that I would never use it again. It had a serious bug that would damage PAL material, causing me search quite extensively for an alternative. I settled for the Panasonic DV codec (although it is only VFW).

Since multi-generation encode/decode cycles were important to me, the panasonic encoder seemed to fare the best in this regard.

It is true that all DV codecs produce the same file format, and that is compatible with all (?) DV recorders - but (as mentioned above) the algorithim used to encode the image is in no way fixed. As a really simple example : 3x4=12 and 2x6=12. The same data is produced both times, but it is arrived at in a completely different way. So the quality of different codecs can be quite significant!

[bb]

Originally posted by mustardman
Having used the Miro$soft DV encoder (spit), I must say that I would never use it again. It had a serious bug that would damage PAL material, causing me search quite extensively for an alternative. I settled for the Panasonic DV codec (although it is only VFW).
This seems strange to me, because as far as I know the Panasonic codec delivers an output being bitwise identical to the M$ DirectShow codec. And I believe I have read somewhere that Panasonic created both codec variants for Microsoft (or vice versa?).

bb

[bb]

As far as I know Premiere uses the standard Microsoft DirectShow filter (not sure, though - I don't use Premiere), and you cannot change that.

bb

[mustardman]

This seems strange to me, because as far as I know the Panasonic codec delivers an output being bitwise identical to the M$ DirectShow codec. Now that is interesting! My experiences are with each codec :

M$ directshow codec accessed within Premiere 5, source interlaced PAL at 720x576 in HuffYUV. After getting really weird results (ie: half the picture shimmered when panning, indicating a field order problem half the image!!] ) I tried a great variety of various things, the one which eliminated the problem was I fed a source image size of 720x577. Anything else (keeping the interlacing) just made it worse.

Panasonic VFW codec accessed within VirtualDub, source interlaced PAL at 720x576 in HuffYUV. No problem at all.

Having to capture at 720x576, add an extra line, then edit & re-encode with the M$ codec was just too much to cope with. The source was video, and I wanted to keep the interlacing. The problem would not occur with non-interlaced source (which it seems most people on this forum deal with - DVDs of cinema release films).

The bitwise identical stream is interesting. How did you come across that information? Although it is not compulsory that VFW and DS codecs are internally different, I thought they would be (apart from just the interface)!?

[edit: Perhaps the M$ problems do not occur with NTSC, after all, that is where M$ is, and they don't really care about the rest of the planet, do they!!!]

[bb]

Originally posted by mustardman
The bitwise identical stream is interesting. How did you come across that information? Although it is not compulsory that VFW and DS codecs are internally different, I thought they would be (apart from just the interface)!?
There were several places where I read about it, but I don't remember all of them. Here's a hint:
http://www.microsoft.com/presspass/press/1997/Apr97/MSMEIpr.asp

Another thread in this very forum:
http://forum.doom9.org/showthread.php?s=&threadid=31741

But these aren't the first places where I got the information. Feel free to search yourself ;)

bb

[Mug Funky]

hmm. i've done a little testing and found that canopus DV codec appears to give the best PSNR when it encodes (overall PSNR ~ 46.5 dB depending on source material), but gives a chroma upsampling bug on decompression.

the problem of b0rked fields given above isn't a feature of premiere :eek: that i know of. had no probs exporting to tape over here (also in PAL land).

if you're worried about using the m$ directshow codec in premiere, you can always enable DV decoding in FFDshow (search here if you don't already have it).

i just stick to the panasonic codec, as i personally feel DV to be a bit of a lost cause as far as quality is concerned. premiere avoids chroma upsampling bugs, and that's all i need to know if i'm working with dv. and it's fast on my slow system.

[mustardman]

@bb : Thanks for the links, I just had a read.

Since the codecs are reputedly the same internally, then some jigging must have gone on with the DS & VFW interfaces? It is interesting that Mug Funky (using PAL) has had no problems. Perhaps it is my install of Premiere that is giving me grief - mind you, I use it pretty infrequently compared to VDub & AVISynth, so I am not particulary worried!

And finally [OT] : DS is supposed to be the "successor" to VFW. I have heard from many software developers that it lacks some very fundamental functions - like frame accurate seeking(?). I have also heard from other programmers that is is a snack to write code for (mind you, they were just playing streams from start to end). What is the real go with DS vs. VFW?