OK, is there a A/B compare of HD1 vs 3chip DV?

[Les Dit]

S-spline PRO does have unsharp mask sharpening.
The 'graphic' preset that was used did indeed do sharpening.
The old s-spline did not have USM, maybe you are confusing the two?
-Les



<<<-- Originally posted by Graeme Nattress : s-spline isn't a sharpener - it's a spline based image upsampler that uses some mightly clever algorithms to make a picture enlargement look a lot better than bicubic.

Also, we're still not comparing the same things - the JVC footage is sharpended to death in camera - nothing can save that in post, whereas the DVX is not over-sharpened.

I'm guessing the JVC has the sharpness always on because the picture can't stand the scrutiny without it. To get colour and luminance of a single chip you can't use all the pixels for resolution - something's got to give. Sharpness gives fake resolution, and is a bit of a band aid for lack of real detail. Before any serious comparisons between the JVC and anything else can be made, we need to see what the JVC looks like without sharpening.

Graeme -->>>

[Graeme Nattress]

Les, you said "No, because you can do the s-spline ( sharpen, really ) to the JVC footage as well."

which makes it sound like S-spline is a sharpening tool, which although it may have some sharpening abilities as part of it's feature set, that's not it's main purpose or ability. It is a very clever upsampler.

As for one-chipedness, yes - if the resolution is high enough you can get away with one chip, but if you want full colour from a single chip of x,y resolution you're only going to get a certain percentage of that resolution as some of the total resolution must be traded in to get colour support.

Graeme

[Barry Green]

<<<-- Originally posted by Ken Hodson :
Can I ask what the time frame is to render each frame with S-Spline Pro? PC or MAC.
Ken -->>>

On my 2.66GHz P4 WinXP single-processor system, it's about six seconds per frame. Obviously a hyper-threaded or dual-processor system may be able to do better.

[Don Berube]

<<<-- Originally posted by Dustin Cross : Can someone shoot a colorchart with a DVX and the JVC side by side and both at 30p? I would like to see one frame from each camera. -->>>

Perhaps we will see some comparisons by Scott Billups in the near future
http://www.pixelmonger.com/hg_cam.html

- don

[Craig Jones]

The original premise of the comparision was that the footage was straight out of the two cameras with no intermediate processing other than scaling. S-Spline Pro does sharpening as part of its scaling process, so it's incorrect to say that the DVX footage is straight out of the camera. Apparently that was always the case. It's irrelevant to argue that the JVC is sharpened in camera. What matters is what the footage looks like out of the camera and what you can do with it. We have a mixed comparison here (though probably as good as can be done). I don't mean to criticize the test but it's important to understand the limitations. What we have here is an honest attempt to improve the DVX footage to HD standards compared to JVC out of the camera. Not all the credit for the DVX footage can be attributed to the camera. If this is a cooked comparision we should edit the JVC footage as well.

Any reduction in full color bandwidth caused by the single chip architecture (read the JVC white paper carefully) doesn't mattter when the format substantially reduces color bandwidth anyway. The whole color bandwidth argument is a red herring for formats that subsample chroma. 3CCD designs have advantages other than that.

I seem to have had a post here lost so I'll repeat part of it. I personally believe the DVX color rendition is hypersaturated and exhibits blooming and color fringing. It's white balance is also different than the JVC. While the JVC could use more saturation, I vastly prefer it to the DVX which I consider awful. S-Spline Pro does a nice job faking the extra resolution on the subject, though. The blue wall detail and the chair give away the resolution differences.

<<<-- Originally posted by Graeme Nattress : Les, you said "No, because you can do the s-spline ( sharpen, really ) to the JVC footage as well."

which makes it sound like S-spline is a sharpening tool, which although it may have some sharpening abilities as part of it's feature set, that's not it's main purpose or ability. It is a very clever upsampler.

As for one-chipedness, yes - if the resolution is high enough you can get away with one chip, but if you want full colour from a single chip of x,y resolution you're only going to get a certain percentage of that resolution as some of the total resolution must be traded in to get colour support.

Graeme -->>>

[Yang Wen]

What do you mean faking the extra resolution? if it looks good to the eye then the resolution is there! Those two words are almost an oxymoron. Since the JVC is already in 720P, there is no need to uprez it using S-Spline. I do'nt see the "cooking" of the test as you refered to. As far as I can tell, it was a simple re-size. You can't judge the colors of the DVX purely by this single example shot. There has been PLENTY of impressive full-rez samples of DVX footage online and not so much good full-rez ones of the JVC. If you want to see examples of how DVX handles color look no other than this guy's website:

http://www.phocinema.com/vietnam/index.htm

Those are perhaps some of the best examples of what Digital camcorders can do. But eithercase... you'd be foolish to say a single CCD video camera can produce colors and latitude better than a 3CCD. Also you pointed out that the walls and the chair gives away the differences between the two cams. Yeah, the chair on the JVC is a total blur! However, if you look at the computer monitor in both shots, you'll see almost the same amount of details, with the JVC perhaps a little better (But that's to be expected). Now I'm not proclaiming the S-Spline is a perfect solution, since the rescaling takes so long.. but on the other hand, so does Magic bullet, and alot of people can't live without it. So I guess if your goal is for a HD projection, then its definitely worth it to spend the extra time at the end to do the s-spline up rez. I can't wait to see what other clever algorithms lies ahead for us.

[Craig Jones]

I mean the extra resolution is faked. It's clearly not in the original camera footage because it cannot be. The extra pixels are interpolated from existing data and then sharpening is applied. The resulting footage is much more than what came from the camera.

Yang, you appear to feel like it's proper to compare two cameras by applying unlimited processing to one and none to the other. More power to you, but I certainly won't rely on your judgement for my opinions. If processing were allowed on both, I'm sure the JVC would have showed better, especially in color. I never said processing was bad, I simply pointed out it was unequal.

I watched all the provided footage on a very high res computer monitor at double size. The chair back in the JVC footage is vastly superior to to the DVX. The back wall is noticably different, some much so that some speculated there was a DOF issue. I don't believe so since the computer monitor was similarly fuzzy in both shots. The digital display and the chair are both in the foreground and there's a noticable difference in quality.

My point about the color is that I didn't like it in the demo shot. That opinion is in contrast to others who've said it was better. If I were processing the footage, I'd adjust the saturation in both shots but in opposite directions. I'd also fix the white balance on the DVX. There were color artifacts in the DVX that I didn't like (the right shoulder, for example).

I didn't say "a single CCD video camera can produce colors and latitude better than a 3CCD". I said that believing 3 CCD's is inherently better than 1 is misguided, especially for subsampled chroma formats. I let the engineers decide how to build a camera then judge its performance. This is in contrast to 3CCD fans who summarily dismiss a camera based on a specsheet. I personally believe that the advantage of 3CCD is in sharpness. That's certainly the case in the still world where there are definite analogs.

[Les Dit]

The dvx footage WAS SHARPENED. The unsharp mask feature was USED in the blowup to 1280 size. How much more clear can that be? I tried it without the UNSHARP MASK sharpen, the results are slim to none as far as getting detail like the JVC cam.


As Craig stated, it's unfair to provide 6 seconds of CPU image processing to each frame of the DVX and none to the JVC. To be fair, the same process should be done to both.
Some might say ' the JVC is at 720p already!'. Well, I say , why limit it to 1280, baby! Bring on 1910.

I did some JVC stills at 1910, and the wonderful s-spline pro can make that look pretty stunning as well.

All that aside, it takes a bunch of work to actually do the S-spline pro uprez-sharpen to the footage. It's not a drop in, turn on plugin. You have to use still frames. It does not work on movie files.
I think S-spline pro, and some of the other advances upres sharpening tools out there are very good tools. Perhaps in the future that will make a NLE plugin as well.

Yes, the sky is still falling on the DV res ( web cam res ) res world. Even if it has 8 chips ;). That's technology.

-Les

[Diu Hai]

Les Dit

why dont you try down-sample your hd1 to sd and sharpen it and compare to dvx to see if the sharpening function does so much different as you claim?

[Kenn Christenson]

Why would you want to sharpen the downsampled HD footage to compare it to the DVX footage? The HD footage already looks considerably sharper than the DVX footage, at least in this example:

http://66.78.26.9/~fiercely/DVXvsJVC/DVXvsJVC.avi

[Craig Jones]

It does. Sharpening, by its very nature, creates detail that isn't already there. Downsampling, then sharpening, proves nothing.

What we have here is upsampling through interpolation followed by sharpening. The upsampling creates higher potential bandwidth in the format. Sharpening fills it in. The added detail is false detail but is pleasing due to the nature in which it's added. The resulting image has higher frequency content that the original. I guarantee it.

I don't understand why this is controversial. No one said the technique is bad, but you can't compare the performance of two cameras that way. People here want to accept this as proof that the DVX has nearly the resolution of the JVC when it does not. It is proof that you can get good image quality at HD res from the DVX through careful upsampling. It is equally true that you can do it to the JVC, like Les said.

[Diu Hai]

why cant compare this way?
dvx can do HD by just some resizing, and its image quality blows the hd1 in HD away, this is done by some software tweaking.
as for the hd1, you cant get a 1ccd look like a 3ccd by color correcting(software tweaking).

[Craig Jones]

I can't see how you arrived at either conclusion. The DVX didn't do HD at all and the JVC wasn't color corrected, so there's no evidence to support either of your claims. As for image quality, there's no doubt in my mind that the JVC looked better in both resolution and color than the DVX.

To answer your question, you can't compare two cameras "straight out of the camera" by heavily post-processing one but not the other. I think that's plain enough.

For all those that claim that the DVX resolution is as good as the JVC, consider the following:

All the detail in the DVX footage is contained with the DV format. If it is the equal of the JVC, that says that all the JVC's detail could be contained in the DV format even though it's in HD. The logical conclusion is that you could downconvert the JVC to DV resolution, then upconvert that DV footage back to HD without losing any detail. Let's see it.

Anyone with the proper equipment could confirm that this not the case without wasting time on it. The HD format has much higher bandwidth potential than DV, so the JVC must not be using any of that if the DVX fans here are correct.

[Ken Hodson]

In my opinion the uprez'ed DVX footage looks good because the talent is very close to the camera. As you can see, no matter what you do you can't pull detail out of a blurry background, no matter what. If you are going to shoot a project where you always have tight shots, this may work, but throw in a wide or long shot and you will see that s-spline isn't going to save your SD resolution.
So far all of the A/B comparisons (Jon's included) have not been done in a way that allows people to see the resolution differance. The focus of the shots have always been a few feet away. Hell my 1chip consummer cam looks great when I shoot close shot.

This is what I would like to see as a definative A/B.
Shoot a person 25-30 feet away, against a wall. This will do away with DOF debates. Center the shot so that the head and foot room of the subject are the same from both cams. Do not worry about width matching unless you are using an anamorphic adaptor. Keep the shots in there origional size. Do not resize or resample. Let us compare a 1280x720 with a 720x480.
And judge!

I think it will be plain to see the resolution advantages of the JVC. No amount of s-spline is going to save the SD footage.
Ken

[Craig Jones]

It's important for the test subject to have detail that challenges both cameras and I think that shooting at a distance will help provide that.