DV500 Preview

[This_Idiot]

The DV500 preview is up.

Check http://idiots-guide.matroxusers.com/dv500preview.html

[DaddyBay]

Thank you for the previews.

The DV500 does solve the biggest problem with the RT2000 (in my view) of marrying the capture card to the video card. I plan on using the capture card far longer than I would like to use the video card. Also, it sure would be nice to have two monitors available through the DualHead plus a TV out like the DV500 setup you saw.

The biggest problem with the DV500 (for my application) is that DV is the only choice for editing. I have tons of old analog footage that I want to save in MPEG-2. It just doesn't make sense to encode to DV and then to MPEG-2, adding an unnecessary recompression.

[This_Idiot]

Hi DaddyB

Yep, it's all a bit of a conundrum isn't it?

I've now got to try and write up a comparison of the two products in the knowledge that a) they are different beasties and b) no matter how I phrase it, either Matrox or Pinnacle are gonna be pissed with me.

So pop in tomorrow and watch me squirm.

[A_BIT]

I have tons of old analog footage that I haven't got around to editing yet. Since I own a Sony Digital 8 camera I plan to just pop the analog tapes into it and capture via firewire, that is when I get a firewire board.
Wanted, cheap Canopus DVRaptor...

-Anthony

[This message has been edited by A_BIT (edited 24 November 1999).]

[Grigory]

DaddyBay,

How can you make mpeg2 from your analog footage now?

I assume you get MJPEG first, and then re-compress to mpeg2, thus "adding one unnecessary recompression." DV format is much better then MJPEG, even with equal data rate.

Next, do you know mpeg2 solution that can capture in final mpeg2 format? How can you edit this video?

RT2000 captures analog video into I frame mpeg2, which is the same by many features as MJPEG or DV. To get final IBP mpeg2 for storage, you always have to "add unnecessary recompression".

I am sorry - this is how our life is designed
You can choose captures of analog to DV with DV 500 or any DV camcorder with analog inputs, and then use any compressor to make MPEG2.
You can capture to MJPEG, MPEG2 I frame, but you always have one final recompression stage to get normal 4-8 mbps mpeg2.
For me, DV format proved to be the best in multiple recompressions, and take reasonable space on a disk. It does not require specialized hardware to do editing: I already have a codec in my camcorder.

The advantage of DV500 vs say Raptor is that it can work without camcorder attached. The ability to digitize analog video to HDD directly seems to be a kind of miracle:
You can do this, but then you have to think how to store this video in DV archive. 14 G per hour can easily fill any HD storage array. You probably take your camcorder and write DV on it. Why not to do this as a first step?

This is what is already possible with any inexpensive DV or D8 camcorder and firewire card.

Chris,
Very nice review, thank you.

Grigory

[DaddyBay]

TI,

I am looking forward to your next installment. Don't worry whether it will be Matrox or Pinnacle who is pissed. No doubt they both will be

Anthony,

I have not used Digital8. I checked on the Sony web sight and while they say that a D8 camcorder will play 8mm and Hi8 analog tapes, it also says (in a footnote):

"Digital Editing on a computer: Videotape acquired on a Digital8 Handycam camcorder is compatible with most types of Digital Video editing applications. Analog video, (i.e. 8mm, Hi8, etc.) must be recorded onto the Digital8 Handycam camcorder before attempting PC based digital editing applications"

It sounds to me like you have to play your analog tapes from a 2nd camcorder/VCR and record them onto your D8 camcorder before you can pass the video on to your PC. I can see that if you want to control the camcorder from the PC this would be necessary. But suppose you just want to capture video from analog tapes on the fly, without camcorder control. Do D8 camcorders only send the analog tape video signal to the analog outputs? Do they also convert to DV and send the signal to the firewire port too? Would this introduce the "shimmy" mentioned in the 25 October thread "Is the internal TBC of Digital8 camcorders utilized during analog footage playback?"

Could one play the tape from a 2nd camcorder and pass the signal into the D8 camcorder and simultaneously pass it over the firewire to the computer?

Grigory,

Yes, capturing to MJPEG and converting to MPEG-2 results in the same number of recompressions. However I presume that if one captures to MPEG2 I-frame using the same parameters that will be used on the finished output video, that the video without effects and transitions will use those same I-frames, unchanged on output. Yes many I-frames will be thrown away in favor of B and P frames, but the remaining I-frames will be the originals (sort of -- there is the conversion from 4:2:2 to 4:2:0 YUV sampling, which is better than the DV path of 4:1:1 to 4:2:0 on output to MPEG-2). It seems to me this would have to result in a cleaner picture.

[A_BIT]

DaddyBay,

The NTSC Sony TRV103 Digital 8 manual says (I am paraphrasing because I don't have it with me at work), "Playback of analog tape will output digital video through the iLink port," or something to that effect. Since I don't have a 1394 card yet, I have not taken advantage of this capability.

I recall reading posts from some who have tried this and said that it does work (without some of the camera control as you mentioned). I suppose the advantage of recording before capture would be a low cost archive of the orignial footage, but then you'd put more wear on the camera, and maybe most of what is on the tape doesn't merrit archival anyway.

I don't think you can pass the analog input through to DV out becuase the camera ports are either going to function as input or output at any given time, but not both at once.

Good question about the shimmy. I wish I had I DV board to test but, I just bought a digital camera and photo printer, so a DV card is not in works for the time being. Maybe someone else can comment on that.

-Anthony

P.S. Thanks for the Pinnacle review. I wonder how much the package will sell for in US$ ??

[This message has been edited by A_BIT (edited 24 November 1999).]

[wkulecz]

Good review!

Is the Marvel tradein offer good in the US?

If so could you post the details.

Dumping my basically useles Marvel towards a DV500 makes it a no-brainer for me over the RT2000.

Glad I managed to hold off on getting that Raptor yesterday!

--wally.

[This_Idiot]

Wally,

I don't have details of US prices for either product yet. As far as the DV500 is concerned, you can take a look at http://www.pinnaclesys.com and check out the special offer. I'm pretty sure it is on offer stateside as well as over here in europe, but I don't know the specifics

[DaddyBay]

Anthony,

Videoguys.com lists the DV500 for US$900.

They list the Raptor for $650 ($500 for the SE version). From what you say, I would conclude that if one's analog footage is on 8mm tape and one is willing to do without the realtime effects, then the Raptor looks like a good alternative to the DV500. The Raptor also has the equivalent of the RT2000's "Infinite Capture" (as I understand it). However, the Minerva Impression CD and DVD authoring software might put the DV500 ahead, depending on one's needs. Then again, the Raptor is a proven product with a loyal following...

[Grigory]

Hi DaddyBay,

You gave another chance to write a "lecture".

1. Using NTSC analog source, like Hi8 camcorder, you have not better than 6:1:1 colorspace equivalent from the very beginning. Please compare the bandwidth of luma and chroma components in analog Hi8 recording standard. What is an advantage of making I frame MPEG with better chroma sampling rate? 4:1:1 will be enough. Anyway,
doing chroma keying with 4:2:2 video made from consumer color-under 6:1:1 tape recording format is not an easy task. You have better ability to adjust keying parameters, but all color contours are still about 5 pixels in width.

2. How do you imagine that it is possible to keep a frame compressed and do editing? When you convert any format to any, the codec has to decompress the frame to RGB bitmap, do modifications and then compress it back.
Original I frame is by most features DCT compressed image as in MJPEG, DV and MPEG formats. Some information is already lost in it. However, something, such as compression/decompression errors, is added. Any codec works with an image itself, but not with DCT coefficients. Do not forget that mpeg codec cannot make P or B frame without decompression of I frame that must remain I frame in output stream. So, making BP frames, the codec will use modified I frame image.
Now, how can you specify to a codec that I frames in original mpeg MUST remain as unchanged DCT data? For this, you must set the codec to use in output stream I frames of exactly the same size as in the input stream. Since the codec has limitation on total bitrate, this may leave too small or too big space for BP frames.
Typically, to get good I mpeg data, you have to set the bitrate about what is used in DV. Remember - both DV and I frame mpeg use very similar compression and the image quality is about the same for the same compression coefficient. This gives you ~120 KB per frame. Typically, the use of B and P frames gives about x2 decrease in data rate. So, using DV-like capture quality you can probably leave I frames intact only if you keep MPEG2 IBP data rate at 1.5-2 Mbytes/sec, or 12-16 mbps. This value is x2 bigger than what you need on the output. The sizes of I, B, P frames must be balanced for many reasons. If you could force the codec to make smaller PB frames, the image quality will change from frame to frame, making the movie picture bad.

So, the codec cannot leave I frame intact, and always have to decompress it for creation of balanced sequence of I, B and P frames. To keep good accuracy, all frames have to be decompressed to image data, and then compressed back. This ensures the best possible optimization of entire movie quality, but not only its I frames.

Have you tried to evaluate movie quality by looking at single frames? In MPEG, you can easily get very good frame images, but entire motion picture will have many mosquitoes and look bad.

Now, what is really helpful, is the ability of some codecs to skip YUV to RGB conversion and back. It is possible now with all codecs: MPEG, MJPEG and DV. All of them use native TV video YUV colorspace. For NTSC they probably have to use YIQ colorspace to be perfectly correct, but YUV and YIQ have better relationship between them in comparison with RGB, so YUV is working too. Even now you can find several compressors, which are able to do recompression without intermediate RGB frame creation. Look at VirtualDub as an example.

However, such ability is not a standard feature of most video editors. All of them now do editing with RGB images inside the effects modules. When the industry switch from RGB colorspace to YUV, the quality of many effects may be improved.

What we finally have: any capture-editing-mpeg2 IBP path involves two compression stages, unless you capture in uncompressed format. The difference between DV (4:1:1), MJPEG (4:2:2) and mpeg2 I (4:1:1 or 4:2:2) is absent from this point of view.

Now about colorspace. The best colorspace is 4:2:2. However, if you capture from analog, the chroma components of 5-6 pixels in line will have very close values, even in the case of 4:2:2 sampling. The image quality will be the same for 4:1:1 and 4:2:2.
While editing, you actually apply effect to RGB image, and resulting RGB frame can have individual chroma values for every pixel, because all calculations are done with RGB values. Making compression to output format, the compressor has to assign individual chroma values to groups of 2 (4:2:2) or 4 (4:1:1) pixels, and only this stage is lossy. Conclusion: in NTSC, the acquisition format choice for Hi8 source can be 4:1:1 OR 4:2:2 with no difference for later editing results. The output format for editing should be ideally 4:2:2, <u>to keep the best ability to use this clip in later production</u>. If you have no plans to edit the clip once again, 4:1:1 format is adequate, because it is what is actually going to analog TV output in any NTSC digital to analog conversion device.

I have left 4:2:0 up to this paragraph. Historically, this format was introduced for JPEG images, containing only one "field", or NORMAL images. When you have single field per frame, 4:2:0 gives you the same or even better quality of image as 4:1:1 - the chroma values are set equal within each 2x2 pixels rectangle. This format is like 4:2:2, but extended to vertical dimension. The chroma resolution is equal in both directions and everything is symmetrical. Chroma keying is done as easy as for 4:2:2, with very good results. You can check this in image editor with JPEG image files. Unless you set too high compression, JPEG images preserve most features of uncompressed RGB pictures.
Do not make fool from a crazy format abbreviation - nothing is "interpolated" as you can sometimes read. The V component IS used as usual, but the 4:x:x formula just cannot describe format. It would be better to take 4x4 pixels, and then count: how many chroma samples are taken for a group. In such notation, 4:2:2 is 16:8:8, 4:1:1 is 16:4:4, and 4:2:0 is also 16:4:4, but with different method. Intel Indeo video raw would be 16:1:1, still not bad for viewing single image.

Why this format is referred to as bad for video? The fundamental reason is that video is INTERLACED. You have two pictures, or fields in a frame. 4:2:0 sampling is applied for each field, but not a frame. So, you get same chroma components values for 2n and (2n+2) lines, and for (2n-1) and (2n+1) lines in a frame, but not simply for n and (n+1) lines. Thus, even for still content in a frame, the chroma resolution in vertical dimension is not 1/2, but 1/3 of luma resolution or less. Furthermore, depending on the method of sampling, you can get for horizontal border of RED object on GREY background the sequence of lines like …RED, RED, GREY, RED, GREY, GREY… with one red line standing separated from actual border. Such distortions are bad for processing and especially for chroma keying. To correct them, you have to smooth out the border by additional filtering. In consumer PAL camcorder’s CCD matrix, such "filtering" is done by the sensor elements arrangement, so you cannot observe separated line on PAL camcorder output in analog or in DV. But you can easily get this effect with PC generated images.
Unfortunately, DV PAL was designed with an idea to provide 4:2… chroma resolution with average number of bits per pixel equal to NTSC. All PAL TV receivers use delay line and display chroma components for each line as [Croma(n)+Croma(n-2)]/2, where n is line number in a frame (remember that lines are going in odd or even sequences, so, in time, the lines with number increments of 2 are received and used in delay line).
The decision to use 4:2:0 sampling in PAL DV was made. Only Panasonic kept 4:1:1 sampling in DVCPRO PAL format.

So, PAL users are tied to 4:2:0 from the very beginning. Conversion from DV to DVD-style mpeg2 is done directly in YUV space and cannot change the quality.
For NTSC, colorspace conversion from 4:1:1 or 4:2:2 in 4:2:0 for mpeg2 has no effect on image quality if it is done only for video distribution purposes. 4:2:0 format is good on screen, but not so good in editing, as I have explained above.

For PAL users, conversion from 4:2:0 to 4:2:2 cannot improve the editing ability, because such conversion only declares chroma samples as individual for each line, but not change their values, initially grouped by line pairs in fields of original video. Doing chroma keying will give equal results on 4:2:0 and 4:2:2 converted from 4:2:0. If the output frames are stored in 4:2:2 format, then you can re-use this video for later editing with minimal losses of chroma keying effects quality. For 4:2:0 output format of edited video, chroma keying problems will grow from generation to generation. This, however, is true for complex editing. If you just <u>insert</u> 4:2:0 clip of any generation AS IS in another video as is , and make 4:2:0 video, you cannot lose quality, because the same pairs of lines will be used in 4:2:0 output chroma samples. Quality deterioration will be evident with moving path transitions, resizing, rotating and other effects. As a result, 4:2:0 video can be recompressed in cut/insert video editing as many times as 4:2:2 or 4:1:1 with similar image quality loss. The statement that 4:2:0 format is bad is true only for "serious" editing operations, and for multigenerational scenarios with serious editing.

For typical home video production, you can:
1. Use ANY of original DV formats for captures.
2. Edit and do any complexity transitions with these video clips. The output quality may be different for 4:2:2, 4:1:1 and 4:2:0 original formats, but will not become better if you do colorspace format "upgrade" prior to editing, such as 4:1:1 to 4:2:2, or 4:2:0 to 4:2:2 conversions. Since 4:2:2 format source video is not available as primary format in consumer DV and analog video, such "upgrade" is useless on capture stage.
3. The quality of edited video <u>on TV screen</u> will be (nearly) indistinguishable for 4:1:1 4:2:2 and 4:2:0 output formats. However, 4:2:2 <u>output</u> format will provide video that is better suited for later editing, but only for those parts of that video, which were produced with complex effects. The parts that contain only data that was copied from source clips to output video maintain the same editing suitability as the original.
4. If you simply compress PAL DV (4:2:0) to mpeg2 4:2:0, algorithm takes Y, U, and V data as separate monochrome images, and compress each of these "image planes" individually. This may introduce compression artefacts, but cannot make further chroma keying worse, because the colorspace itself remains the same, and there is no chroma re-sampling. Compression itself may give you troubles in chroma keying, but it is common for any internal colorspace format.
5. There is no reason to convert PAL DV to mpeg2 4:2:2 before editing. You can get some post-editing (or second generation editing) capability improvement only if your editor do the following: a) decompress 4:2:0 inside editor to 4:2:2 or 4:4:4 (or RGB), b) apply effect that modifies chroma values for pixels <u>individually</u> in 4:2:2 or 4:4:4 colorspace, c) compress this uncompressed frame to mpeg2 or mjpeg 4:2:2. If b condition is not followed, for example, you apply simple cross dissolve effect, you further ability to use this video in editing is not altered. Filtering like changing color balance, brightness, contrast, wipes, iris transitions, are all falling in this category. All resizing, smoothing, moving path, 3D transitions may benefit from 4:2:2 output format, but only for further use in chroma keying and similar operations.
6. If you have no plans to use your output video in later editing, later editing of this edited video, later editing of this edited video, …, …, there is no advantage of 4:2:2.

Hope this is enough for today .

Grigory

[DaddyBay]

Grigory,

Thank you for the lecture. I thoroughly enjoyed it. All frames are recompressed. I stand corrected.

I take it that in your opinion using either the RT2000 (MPEG-2) or the DV500 (DV) for editing analog Hi8 footage will result in about the same output quality, assuming the video is edited just once.

I am curious about your opinion of Matrox's assertion that the "RT2000 features uncompressed, 32-bit, 4:2:2:4 graphics for maximum quality. DV500 uses DV compression on
graphics that reduces the resolution to 4:1:1 and can produce significant artifacts."

[Grigory]

Hi DaddyBay,

This is correct, if the footage originates from 4:2:2 format, for example, you make animation on PC as RGB uncompressed movie.
If you have consumer analog video, or NTSC DV video, you already have 4:1:1 or worse, so making oversampling may have only minor effect on quality or none.

You see, it is "better" to listen 100-7,000 Hz @ 2% distortions AM radio on 0-20,000 Hz amplifier with 0.00001% distortions, but can you detect a difference?

I still cannot understand how is it possible to get as good as DV quality with much less bits per second.

What Matrox says, can be achieved by conversion of DV into RGB (4:4:4) - but how can you make the quality better? Mr. Marketing knows. This person always knows better

I tried to figure cases when 4:2:2 is really good. They simply do not work for 1-2 generations, which is my limit of editing.

Note: I am not saying RT2000 is bad, I just say that we should not expect huge improvement in video quality with this card, compared to pure DV.

Regards,
Grigory

[wkulecz]

Thanks for the DV500 link. Unfortunately the special offer doesn't seem to be on the US site.

Looks like the PALers finally get something first :-)

--wally.