Capture Quality 1394 vs MJPEG

[Frank Marshall]

Wow! And here I've been suffering from a mild case DV-envy all this time!

Just out of curiosity, do the DVCAM and Panasonic's equivalent, what's it called, DVPro(?) formats outdo Hi8 in this regard? Or does it take Digital-S (DV50) to match Hi8 for MPEG encoding?

[Ted222]

Doc,
Now that you have mentioned the RT2x00, how much better are the results? As good as Mjpeg through the S-Video capture?
Ted

[johnpr98]

JerryH

<font face="Verdana, Arial, Helvetica" size="2">
I have been seeing the artifact stuff to which you refer (I have a D8 camera and Pyro firewire card) after rendering</font>

Just checking to see if you have upgraded to DirectX 8.0 with the newest version of qdv.dll or tried Main Concept

John Price

http://www.johnpr98.com




[This message has been edited by johnpr98 (edited 20 June 2001).]

[Dr Mordrid]

The RT's output is excellent. The realtime effects, including 60 of Premieres built-ins, are done with very high quality. Much better than software rendering of the same effects. This is again a function of its compositing engine. Output to analog devices is also excellent.

My biggest problem with it is choosing an appropriate option amongst the 1700+ effects when I use one.

Of course a good editor only uses <5% special F/X in normal editing, exclusive of start and end titles. The remainder should be cuts and dissolves.

Contrary to what a lot of folks think much of the RT-2x00's advantage is during the editing process itself. When you edit a lot much of your time is spent waiting for previews to be generated before they are reviewed. With the RT's this is instant.

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 20 June 2001).]

[Frank Marshall]

floggin' a dead horse?

Sorry if it seems like I'm floggin' a dead horse but I really want to get this right. I quizzed Adam Wilt, the DV guru, about whether Hi8 or DV are better for encoding to MPEG. Here's our correspondence in its entirity. I hope I didn't overly abbreviate your case, Doc. I'd like to get to the bottom of all this. Thank you:

> There's an interesting thread at the Matrox users forum regarding Hi8 vs DV.
> Doc Mordrid, the resident guru, claims that Hi8 footage (captured with the
> huffYUV codec) encodes better to MPEG than DV. He puts this down to the DCT
> and colourspace limitations of DV. Could you please tell me whether you
> agree with this contention or not? Here is the URL for the thread I've
> mentioned:

I'm sorry, I don't have time to read this thread.

I have not encoded Hi8 to MPEG-2 for DVD. I have seen DV transcoded and put
directly on DVD. It looks great.

Hi8's colourspace is more limited than DV's, both in "sample rate" and in
gamut. Furthermore Hi8's inherent luma and chroma noise and considerable
timebase instability (even with the use of a TBC) add unwanted complexity to
the image that makes for difficult coding. And I say that as a five-year user
of Hi8 in industrial and professional settings.

If capture to the huffYUV codec looks so good with Hi8, what happens if you
capture DV to that format? It may be that the filtering used in the capture
process (I'm guessing; I don't know the codec or the capture card) is cleaning
up the image sufficiently to reduce the load on the codec. If so, the DV image
should look even better than the Hi8.

Cheers,
AJW

[Dr Mordrid]

The key is that he himself said he hasn't encoded Hi8 captured in any format and encoded it to MPEG, much less pure YUV or HuffYUV, so he's making a comparison based on presumtion, assumption and visual impression.

It takes a lot more than a though experiment to get good MPEG's.

First, I find his points of reference interesing. Hi8 isn't a video format like DV25 is, it's a tape format. The video format is YUV @ 4:2:2.

Also, a lot of what I'm avoiding DV25 for is studio shots involving effects (compositing, greenscreen etc). In those shots there is NO tape involved as I'm capturing the S-Video passthrough directly to the card as a direct feed. Therefore, no tape losses or noise.

I do also tape on Hi8 when filming real world subjects that I KNOW will be a problem for DV, such as those with strong diagonal components, sudden brightness changes, fast motion etc. that are known to trip up DV.

In these situations the tape noise I'd get on a new Hi8 tape would be the least of my problems compared to what DV would do.

So, on to the rest......

"Hi8's colourspace is more limited than DV's, both in "sample rate" and in gamut."

Judge for yourself;

D1 4:4:4 (full sample rates for luma and chroma);

Lumance: 13.5 mhz
Chroma: 13.5 mhz

YUV/YCrCb 4:2:2 (Hi8's S-Video output);

Lumance: 13.5 mhz
Chroma: 6.75 mhz

DV25 @ 4:1:1;

Lumance: 13.5 mhz
Chroma: 3.375 mhz

Summary: YUV is sampled at 1/2 the D1 rate; DV at 1/4 the D1 rate. So where's DV superior in this regard? Got me.

The only thing I can think of is that in DV the luma is PROPORTIONALLY a higher percentage of the total signal. This results in a higher contrast and seemingly sharper image that pleases the eye since the human eye is more sensitive to brightness than color. In short, it's an optical illusion that explains why DV "looks great" to casual observers and IS fine for "natural" video.

While this may please the eye it's not always the best thing when transcoding. The eye doesn't concatenate the compressed video data, encoders do. Concatenating compressed video data that is already sampled at a low rate does not much good.

In terms of gamut (I'm assuming he's referring to NTSC/PAL filtering) digital does have a slight advantage, but nothing that can't be made up for OR relied upon. In fact I've run into problems in this regard using Sony D8, Canon DV and Panasonic DV cams, mostly related to white balance inaccuracies. DV isn't perfect at this *in the real world* by any means. Manual white balance can help, sometimes a lot, but it's not a cure all.

I'm sure that like many pros he's mainly seen DV encoded to 720x480 MPEG-2 for DVD at moderately high bitrates. With a large frame size works because most errors will be small enough in proportion to the whole frame to not cause many issues, but that's what typical users are doing is it? Nope....

When encoded & scaled to smaller frame sizes, as for VCD or SVCD, the concatenation, scaling & DCT errors pile up and become objectionalble with any compressed format, DV especially because of its low color sampling rate. Objectionable enough that the problems overcome the codecs strenghs and affect final quality. Start out with a reduced sample rate and artifacts can get out of hand fast.

Dr. Mordrid

[jeff b]

It should be noted, however, that even though the DV format and colorspace could have been better thought out before becoming such a big hit with consumers, it is, nonetheless, the best thing since sliced bread for capturing, editing, and outputting back to DV or VHS, when all you're really talking about is consumer level "home movies", etc.

It certainly doesn't lend itself to further compression for streaming video, etc.

Jeff B

[Dr Mordrid]

I totally agree. For 90% of video work it's just fine. It's for those little exceptions that it trips over itself.

I'm assuming you're talking about it doing its Huffman encoding at the wrong time?

Dr. Mordrid

[Vidgal]

Great! I just found a DV camcorder in my budget and it's on it's way, but in order to offset the costs, I'd planned to sell my Hi8. Sounds like I should keep it. In fact, it sounds like I should have stayed there in the first place. I knew there were some compositing limitations, but I've never seen it discussed in so much detail or with the caveat that Hi8 might actually be better. If DV has all these problems, why is everyone and everything leading there? I feel very confused and disappointed. I was hoping my new camcorder would give me better quality final output, not worse!

If I output from the new cam with S-video to the G400, will this help or do I need to keep the Hi8 and shoot totally separate footage for re-encoding use.

[Dr Mordrid]

People are using DV for several good reasons;

1. frame accurate device control

2. good enough quality for NATURAL video, which is most of what folks do

3. the cams are generally lighter and smaller

4. the ability to do scenealyzer type captures

5. no quality loss transferring video into the computer and back, with the exception of those losses that occur in EDITING.

On the other hand low (or no) compression analog captures give you;

1. a larger colorspace, which helps prevent stair-step artifacts in composites

2. a softer look, which also helps when compositing. Sharp & contrasty isn't always the way to joy.

3. much less artifacting on recompression or transcoding to other formats. With YUV or HuffYUV there are no DCT blocks to cause artifacts.

4. fewer or no (depending on the codec) motion, stairstep or DCT block artifacts of the types you get in DV. Codec issues again.

The uppance is that YUV or low compression is better for UN-natural video, for lack of a better term, and for those situations where DV is likely to cause trouble. These can be painful and are why I use the Hi8 when I see them.

IF I had a Betacam I'd be using that instead of the Hi8, but it's not in my budget

Terry

[bogie]

Hi everybody!

just a small correction on Dr.Mordrid's "color space"...
the 4:4:4 sampling scheme is always used for the RGB space.

bogie

my home rig:
epox KP6-BS, celeron600@900, PQI 512MB,
G400DH vanilla@170/170, flashpoint SCSI,
cheetah9.1 + 2XquantumXP4300, SBLive value,
realtek8139 NIC, generic BT878 video capture,
philips brilliance17A + sony GDM1936

[Dr Mordrid]

I wasn't talking RGB but YUV.

But, since you brought it up most times RGB video is mathematically derived from YUV and actually exhibits some losses as a result.

This is why when given the choice, as on Matrox caputre cards, you should choose YUV/YUY2 etc.

Dr. Mordrid

[aaroncgi]

Man, where was this thread a couple weeks ago when I started looking at DV cams! I haven't bought one yet, so all is not lost, but if I have to go out and spend hours on the web and stores again trying out every camera..I hate shopping, I like buying.

Anyhoo, was all set on a Canon Mini-DV cam, now this thread had me thinking twice. Though primarily, my video will be used for transfering to VHS and watching on TV, I also would like to post them on the web. No idea on window size - it will depend on the memory required. Most of my friends and family who'd be downloading and watching them are, like myself, stuck on a 56k dial-up connetion. As such, it's almost a given than any video that you can download in less than 5 minutes (and many that take much longer) are going to be small and grainy and just not look very good.

Dr. Mordrid, really appreciate all the thoughts on DV vs Hi8 et al. I may have read it wrong, but it sounded as if you might have suggested that DV video CAN end up looking good, if you use the Matrox RT2x00 to convert the raw footage to your MPG, AVI, etc.. (I will probably do MPG). Of course, the RT2000 is $900 last time I checked, and the Mini-DV cams are naturally more than the Hi8 for the most part. So, this would no doubt be the expensive way to go.. But if I go 8mm, then I HAVE to get a good video capture card, can't just stick in a decent IEEE1394 card.

My other concern is how the Hi8 would look on a good television vs. Mini-DV (playing straight from the camcorder). All indications have lead me to believe that the Mini-DV cams will have noticeably superior picture as compared to the Hi8, although this I think should be nullified if you copied one of each onto standard VHS, right (due to the limited resolution) ??

Do I have any grasp on this stuff, or no clue what I'm talking about?

Aaron

[Dr Mordrid]

hehehe......complex topic, yes?

Actually you can get decent DV => MPEG using the RT-2000 because it processes the video at 4:4:4:4 RGBA. This is what gets fed to the encoder, be it a plugin OR by frameserving TMPGEnc using AVISynth.

Yes, the RT-2000's use full quality RGB internally with an alpha channel for the overlays. This is why the RT-2000's effects composite so clean. That and a very good DV codec in the output stage. MUCH better than MSDV, that's for sure.

Still there is room for improvement as doing bluescreen and other such things are still problematic in the RT-2000 because of the colorspace issues. Alpha channel is one thing, blue/greenscreening and keying are another bird altogether. This is where the analog gear gets heavily used.

YUV @4:2:2 AND DV both have their place. Each does something right. IMHO the job now is deciding where to use what. NEITHER is perfect for everything.

Now to get the @#&% manufacturers to produce a properly done 4:2:2 NON-TEMPORAL digital format to replace DV25 at reasonable prices. Nope, I'm not hopeful for MPEG cams.

This has been done at the pro level with D9, DVCPRO50 and HDCAM, but at $12,000-100,000 a copy.....

Dr. Mordrid

[aaroncgi]

Ok Dr., well most of that stuff went right over my head, but thanks! Sounds like your saying mini-DV isn't all evil, so that's a start. And I can't afford a $12,000+ camcorder either, the one I am looking at is $500! I realize this is a Matrox oriented forum, but is there any product out there that's even close to the RT2000 in terms of image quality when creating mpegs from a mini-DV, but doesn't cost $900?

The camera I'm looking at getting comes with a free Intro DV card/software offer, I have little idea how capable it is..

Aaron