Both clips, up-rezzed from the camera original up to 1920 x 1080. The DVX clip is 1920 x 1080 x 24P, the JVC clip is 1920 x 1080 x 30P.

http://66.78.26.9/~fiercely/DVXvsJVC/DVX-1080.mpg

http://66.78.26.9/~fiercely/DVXvsJVC/JVC-1080.mpg

The new DVX 1080i clip looks better than the old 720p clip. Shouldn't it look worse?
Ken.

The new DXV clip looks very soft to me. It also suffers from a motion artifacting due to the scale -- you can tell the subject is moving in a lower resolution space. This is clearly more detail in the JVC footage, as I converted them both to CFHD AVIs and found the bit-rate climbed 20% for the JVC over the DVX footage -- this is a quick and dirty non-subjective experiment that messures detail.

found the bit-rate climbed 20% for the JVC over the DVX footage

Keep in mind that the JVC was 30P, the DVX was 24P. That alone should account for the 20% increase in size.

I've got more clips finally, which I'll be posting later. I've been following this argument back and forth because I find it interesting, plus by posting the clips I pretty much started it. I think people have taken it on tangents that I never expected, which is always fascinating.

The original purpose of posting the up-rezzed clips was to see how the cameras compared at comparable resolutions (both by up-rezzing and downrezzing). But a lot of discussion veered off on color rendition, and how some people liked the DVX colors so much more, whereas others think they're super-saturated unrealistic, and that the JVC has the more accurate color rendition.

For the record, the DVX clip shows the color that was in the scene much more accurately. It was hyper-saturated, with lots of purple and blue gel light thrown in. But the argument about color is completely specious and should be put to rest: the DVX can deliver any kind of color you like. If you don't like what you're seeing, keep in mind that the DVX gives you control over detail level, chroma level, chroma phase, color temp, master pedestal, gamma curve, skin tone detail, adjustable color matrices, and the DVX100A gives you even more control. You can dial in exactly what color you want, so color rendition should never be used as an argument against the DVX, only against the JVC, since with the JVC you get what you get whether you like it or not.

After extensive use of both these cameras, and especially after discovering just what S-Spline Pro is capable of, I have come to the following extreme analogy to describe what the choice is like (to me, others will disagree):

Let's say you have two doors to choose from. Behind door number one, you will be given an expert hour-long Swedish massage by a gorgeous and talented masseuse, after which you will be handed $80,000 in universal currency that can be spent anywhere in the world.

Or, behind door number two, you will be given $100,000 in Sears gift certificates, good only in Sears stores in Kansas.

That's how I see it. Using the DVX is luxurious. Using S-Spline Pro to up-rez, you'll get about 80% of the picture that the JVC is capable of. However, you'll be able to use that picture anywhere: you can transfer to film and show it in any theater in the world, you can broadcast it on HDTV, you could use the SD signal on NTSC tv or easily convert it to PAL and have a worldwide broadcast. By comparison, using the JVC is relatively torturous - no picture controls, no true manual controls, no remote jack, no ND filters, no monitoring of HD as you record, no focus marks, and on and on. You'll end up with a picture with higher resolution, yes. But it can only be used in a very few certain areas. It cannot be transferred to film, it cannot be transferred to PAL, it can only be theatrically projected in areas that have digital projectors (probably less than 1/100 of 1% of the theaters in the world), it can only be received in its full HD glory by about 1% of the US population (those that have both HD TV's and an HD delivery system, meaning an HD tuner, HD cable service, or a DVHS deck).

A less controversial analogy would be this: the JVC is like using a 35mm Cameflex to shoot 35mm Ektachrome color reversal film. The DVX is akin to shooting Super16 Vision 2 color negative, using an Arri SRIII. It's actually a fairly accurate analogy because blowing up S16 to 35mm requires about a 3x optical magnification, about the same as SD to 720P HD. And using today's digital intermediate technology is equivalent to using S-Spline Pro. S16 negative will give you exceptional control over the image, whereas 35mm will unquestionably have higher resolution, but the Ektachrome will have much narrower latitude, etc.

To each his own, use whatever tool you prefer. For my purposes, and I'm sure for many others, the JVC's increase in resolution is just not worth the compromises, but for many of you that may not be an issue. At least by posting these clips, you now have more of a basis for comparison to base your choice on.

Once the next generation of HDV cameras arrive I expect that my position on this comparison may take a 180-degree turn. An HDV PD170 or XL2 or DV5000U should (should) smoke the DVX in all measurable ways, not just in resolution.

<<<-- Originally posted by Barry Green : Keep in mind that the JVC was 30P, the DVX was 24P. That alone should account for the 20% increase in size.
-->>>

I had already taken that into account (remember I do this for a living -- video compression technologies.) The total data size increase was 40%. Only 20% was due to increased detail in the JVC footage, the other 20% like you suggest is the different between 30p and 24p.

While the luxury vs. abuse argument certain has a ring of truth to it, the universal currency vs. Sears gift certificate argument does not. You're suggesting that the JVC footage can be used for virtually nothing while the DVX can be used for everything. That's preposterous. The DVX offers 24 frame progressive but is otherwise just an anolog TV format. It's HD (as if it offered HD) is not competitive with the JVC. Both formats can be transcoded to other formats with varying degrees of success. The DVX is just as far from the universal camera as the JVC is (neither comes close, in other words).

Being a software developer, I'd much rather take on the task of downconverting 30 frame to 24 than upconverting NTSC to HD. One increases resolution while the other decreases it. It may be true that no good software exists for that conversion but that's not proof that it can't be done. I note that the software used in the comparison is not specifically made for video processing. Still frame upscaling is something people have needed for a while so it's no surprise it can be done. Doesn't deal with motion artifacts though.

I have no doubt that future HDV cameras will be better than existing ones but that's not what this argument is really about. People don't like that the camera is from JVC and are eager to declare that imaginary cameras from preferred manufacturers will "smoke the DVX in all measurable ways, not just in resolution." I'm sure that future cameras from JVC will as well. So what? Hiistory may show the HD1 to be the worst HD camera ever, but that changes nothing today.

The JVC is a pioneering camera in an upcoming format. You may choose to use it with all its present shortcomings or you may not. When HD cameras become easy enough for even Sony to build, everyone will offer one. Until then, DV resolution is still inferior.

<<<-- Originally posted by David Newman :
I had already taken that into account (remember I do this for a living -- video compression technologies.) The total data size increase was 40%. Only 20% was due to increased detail in the JVC footage, the other 20% like you suggest is the different between 30p and 24p. -->>>

Doh! You're absolutely right, of course. Thanks for your restraint in your gentle correction!

No question the JVC is much more detailed than the DVX, I have always given it its props for that.

Dealing with temporal changes (in this case 30p to 24p or 25p for PAL) is a few orders of complexity harder than changing the resolution of the picture. This is because in the temporal resolution of all video formats are low - ranging from 24 to 60 depending, whereas video resolutions start at an order of magnitude higher at 480 lines, and go upwards from there. A basic bit of number theory will show why a good standards conversion is hard, and why motion estimation is widely used technique.

From this, it can be seen that 24p is perhaps the most universal format there is, whereas 30p is the least universal format. 24p converts to NTSC (60i via 3:2 pulldown) and film, and PAL (by speeding up 4%) all with most excellent quality. No magic motion estimate is needed because the numbers divide out nice.

To convert 30p to NTSC (60i) is fine, but to convert to PAL will require some compromises to quality, as will a conversion to 24p. It's not a straight conversion and losses will occur.

What we really seeing is that resolution isn't everything. I personally prefer the look of the Panavision HD cameras, even though they have less resolution than compared to film. I personally prefer the look of a top end DV camera to the HD1, which looks like high resolution VHS to me. Sure it has more resolution than DV, but I want quality, not quantity! I'd rather have 480 good looking lines than 720 bad looking ones. But that's my personal choice. As soon as 720 good looking lines are affordably available, then that will be my new preference.

Remember a low end DV camera will have a higher resolution than a high end Beta SP camera, but the BetaSP camera will produce a better picture, lack of resolution not withstanding.

Yeah, and it's all hard when you have nothing to do it with, too. I'm not sure what "orders of complexity" are or what number theory you are talking about. Are you suggesting that smooth variable speed is an impossible task? Surely interpolating frames is not orders of magnitude more difficult that recovering detail that's gone and lost forever.

I've got a box in my living room that does a great job of converting 60i into 60p (in fact, it's a quadrupler). Once upon a time I'm sure that was considered really hard as well. These days there are realtime devices that deinterlace, do reverse 3:2 pulldown, and arbitrarily scale video all with really high quality. That sounds like a lot of "orders of complexity". If 30p to 24p film conversion isn't done easily it's because there's little demand. I don't believe it's an impossible task. It doesn't concern me in any event, since I have no interest in output to film. This is a prosumer camcorder forum after all.

Apparently low resolution frames and low resolution time is the holy grail for some people, but I don't think 24 frame film conversion is the desire of all people and certainly not all those interested in the JVC. For those that seek low temporal and spatial resolution, enjoy your DVXes.

The difficulaty of 30p to 24p has nothing to do with demand, and everything to do with basic mathematics. It's certainly not an impossible task - who said it was? It's just that certain frame rate conversions produce better for results for equal effort, and some conversions produce poor results for a lot of effort and processing. Nobdy doubts that 24p to 60i NTSC produces pretty good result by 3:2 pulldown. It's a fairly mechanical and deterministic algorithm as you well know. The algorithm for 30p to 24p is neither deterministic nor simple as it necisitates motion estimation. Normal video runs at 60i, which gives any temporal processing twice as much information to go on, and a fairly simple algorithm to go to 24p, whereas 30p doesn't. It's a matter of numbers.

Why do you think PAL people put up with a 4% speed increase from film conversions? It's because the alternatives didn't used to exist (motion estimation is relatively new) and because of 24 and 25 being relatively prime, would necessitate every frame being processed, reducing the image quality. The numbers dictate the ease and quality of conversion.

30p to 60i is easy - it's a factor of 2 involved.

As for having nothing to do with it, I've written standards converters, so I certainly have both the practical and theoretical experience to know wether certain conversions are easy or hard, and the relative qualities of them.

As for people liking 24p, that's fine for them. Certainly 60i or 50i produces very nice smooth results for video, and for modern TV, 60p or 50p would produce superb results and resolution, but film is going to be around with us for a while, and we're pretty much stuck with it's 24p nature.

As for the resolution debate, it's pretty obvious that the JVC represents a trade off for resolution over quality - how else could it be the case on a camera that's the first of it's kind for the price it sells at? If you agree with that trade off, then the camera is for you, but I think the rest of us will either wait for less compromised cameras, or save up for the real HD gear that's available.

The JVC almost certainly represents a view of the future, but for most of us, the future isn't quite here yet.

Graeme

This continuous battle between the DVX and the HD10 invariably hangs on one overly subjective issue: COLOR.

For those of us who may have no interest whatsoever in color rendition, how does this topic shift when the comparison is over black & white footage?

Has anyone posted and A/B comparison between these two capture tools displaying their output in either black and white mode, or where color is removed in post?

Such a comparison would surely remove the most subjective part of any A/B analysis, and would answer what is for me the key question: "Just how well does each camera serve as a replacement for film?

[Also, since the DVX100 is often compared against the HD10, and the HD10 is often compared against the Cinealtas and Varicams, the question is begged, does anyone dare compare the DVX100 against the Cinealtas or Varicams?]

Brian

Graeme,

I'm not suggesting that any arbitrary framerate conversion is as easy as things like 24p or 30p to 60i, what I'm saying is that it is not a more difficult problem than trying to upconvert DV to HD without artifacts.

The claim was that the DVX was useful for both film and HD because of 24p and upconversion, whereas the JVC couldn't be used for film because it did not offer 24p. My statement was simply that I'd rather try to solve the framerate conversion problem than the spatial artifacting problem. The difference apparently is that spatial upconversion tools are more readily available that 30p -> 24p conversion tools.

You may have theoretical and practical experience with these matters but you haven't offered any. Instead you've talked about "orders of complexity" and "number theory". I'd like to remind everyone that number theory is a branch of mathmatics that concerns itself with the nature of numbers and is, of course, irrelevant to the conversation. Please explain how arbitrary framerate conversion is "orders of complexity" harder than arbitrary spatial upscaling. My professional background is in electrical engineering, computer science, signal processing and firmware development. I may not have done this specifically, but I certainly understand what's involved.

You've suggested that the HDV format will only be proven once someone other than JVC makes an HDV camera ("I guess we'll only find out what the HDV format is capable of when it gets used in a professional camera, rather than the JVC.") and also suggested that the JVC sharpness issues are due to an inadequate lens ("I would then have to say that whatever lens you have on a DV camera, you'd need a better one on a 720p camera to do any justice to the resolution. Perhaps that's why the HD1 always has the sharpness set too high...."). Perhaps you can be more objective regarding these conversion issues.

Some enthusiastic threads are around where people said they were going to post some such footage comparisons, but inevitably they never got that far. Just lots of eloquent rhetoric, touchy feely almost. Perhaps I exaggerate just a tad, but still, no images.
Well, I did notice that the www.pixelmonger.com guy did have a few resolution charts from the high end HD cameras, and he mentioned the HD10, but no image from that.
So here is what I am doing: I took the res pattern image :
http://www.pixelmonger.com/hd_assets/cam27V.jpg
from the Panasonic camera, and I re-rendered the right half of the charts res pattern so it's totally 'clean'. I then printed the chart on a 11"x17" high quality inkjet. So now I have a nice chart to try with my HD10. The chart to compare with is the original that I linked to above.
I don't think it will look as clean as the Panasonic, but it should be interesting.
-Les

The "orders of complexity" for frame-rate conversion difficulty is pretty true, if you do it the absolute 'best' way.
The technique used by some of the better programs is based on optical flow. I dabbled in it a bit, and had my programmer friend port some code from an SGi to PC for doing it. Basically, optical flow is motion tracking , using a resolution pyramid. It tried to produce a flow vector field for all motion between frames, which then can be used to synthesize frames 'inbetween' any two real frames. That's how you make the new frames at the new sample rate ( 24 ). The process is computationally expensive, and prone to all kinds of errors when objects turn and reveal new information that wasn't there in the previous frame. I'm not sure about "orders of complexity" harder, but it is many many times harder than just spacial upresing problems. The code makes S-spline algorithms look easy.

But here is something interesting, I've looked at footage that was converted to 24p just within VEGAS's re-speed, and it didn't look objectionable to me at all. It looked very normal, and I tested some footage of some skaters with a lot of motion as well. Maybe it's much ado about nothing, I dunno.
-Les

"much ado about nothing" -- well worth trying, though!

I've seen 30p footage converted to film, and it was completely unwatchable, the motion artifacting was absolutely objectionable.

But that was a few years ago.

We could certainly give today's tools a try. When I do the next round of split-screen JVC/DVX footage (some outdoors material, plus some scenes from our "matrix" commercial) I'll also just give old Vegas a run at outputting a 30P file as 24P and see what it does...

I don't know about the specific alogrithm you describe, Les, but I'm familiar with the things that editing packages can do and know they are of varying quality. I've never specifically tried 24p conversion since it isn't of interest to me. It is clear that the problem is much more than simply interpolating pixels between frames.

On the spatial end, the upconverting Barry did exhibited motion artifacting as well. Not really fair since the tool wasn't made for motion, but still, full motion spatial upres'ing is a difficult problem, too, if you do it the "best" way (that being something much more motion-aware than S-spline Pro).

I think the issue is not just which one is harder but which one is more objectionable. Frankly I don't see the problems as all that different but I do believe there's a difference in sensitivity to the two problems. How would you eliminate spatial motion artifacts without using motion estimation?

David,

Thanks for your contributions, we really appreciate it. You take the second guessing out of a lot of things. How about a Final Cut Pro plug-in?

heath

Hi all,
I know I'm a bit late in this discussion... but since all links
to footage doesn't work anymore, can someone re-post
links to available HD10 footage?

I'm of the ones of think that 1CCD-HDV is better than 3CCD-DV also
in terms of color quality, I've developed some testing tools to validate this hypotesis but I cant do the test since I dont have the same Footage from both cameras... could some one provide it?


thanks
Alarik

I have an HD-10U. Any particular type of shot you would like to have?

Hi Robert,
thanks for replying.
If you can I'd like just few second of an exterior (a park would be great)
on a sunny day, someone walking and static background in 720P.

Thanks a lot
Alarik

It's the dead of winter here in Minnesota and we're getting some very frigid temps coming in. I'll see what I can manage for you.

Thank you very much Robert,
any exterior will be great!
Thanks again
bye

I'd question the point of comparing an early 1-chip HDV camcorder to one of the most mature 3-chip DV models, which currently sells for maybe twice the price. A more relevant test would be the Sony FX1 versus the DVX100, and I shot some footage this week on these two cameras plus several others which will be available soon on the WEVA forums. From what I could see on the viewfinders the DVX100 appears to produce a pleasing SD image, but in general I've yet to see any SD footage which holds up well when compared to decent HD footage on an HDTV display.

For SD delivery it's a more level playing field, but even that gets tricky when you consider widescreen vs. 4x3 issues. The DVX100 I used this week had an anamorphic lens adapter on it, which may be a useful way to extend the life of SD cameras as we head into the HD era. But if you're thinking of buying a new camera today, why not consider the HVX200 instead of the DVX100? With the HVX you get widescreen DV recording to miniDV tape and the potential for high-quality HD recording in the future, once the cost of recording options for the HVX comes down to reasonable levels.

At today's going prices, used, an HD10U would be quite viable for acquiring 16:9 (in good lighting) to downconvert to SD, IF the HD10U offered reasonable control of the image. It will almost certainly outperform (again, in good light) a 3 chipper that uses sub 400k pixel resolution 4:3 chips without an anamorphic lens.

You are right on the money Robert. I whole heartedly agree.