4:4:4 12-bit Uncompressed DVX100

[Jason Rodriguez]

Theoretically there should be a huge gain in dynamic range, at least 2 stops more. The reason?

With the current DVX100, you are dealing in post with a gamma-corrected, 8-bit, 5:1, 4:1:1 compressed DV Signal. If you ever try to underexpose your DVX and then "bring up the shadows", forget it, unless you like the look of those square pixelated compression arifacts lurking in the shadows. With 12-bits and the number of levels it gives you (4096), there is a lot more room for underexposure to bring the highlights down and under control, but with enough information in the shadows to then bring them back up without compression artifacts or banding.

Seems to me that it would end up being similar or the same to what the camera normally offers, because I would guess that the camera just has a set curve/gamma adjustment it makes on the image before it gets compressed and saved to the tape. If that was the case, it would seem like if you used a similar setting, that you could expose properly for the tape and the firewire out...

Yes, that will be the case, but the nice thing is, you don't HAVE TO expose like you did with DV. You now have the freedom to underexpose the picture to keep the highlights under control, but still have plenty of information in the shadows to increase the color saturation down there, and also normalize them/apply gamma, etc. without banding or other compression artifacts poking their heads up. That's why I'm saying at least 2 if not 3 more stops, depending on the amount of noise in the shadows. Now your only limiting factor in playing/color-correcting the picture is noise, not extreme compression and it's nasty side effects out of the current DV stream from the DVX.

[Joel Corkin]

So, how about some proper tests, Juan? It's not anything that any of us can do, even from images you post, unless you post a whole slew of DV and uncompressed images side by side. You can use a single set up either outdoors or indoors, and just run the gamut of full stops the camera has, from next to fully closed to fully open, while noting all the fstop settings of each frame you capture.

We will be able to see on what frames you just start to see shadow detail and on what frames you lose all your highlight detail. The lighting has to be constant and should be within a range that is useful for the test. Some experimentation should help figure out what that is.

hey everyone, i did some quick tests to see roughly what the potential latitude might be. the info and images are at
http://www.jaanshen.com/juanmod

though unscientific, they're very exciting (at least for me).

[Jason Rodriguez]

BTW, with anything more than 1000:1 (10 f-stops) and 12-bit linear, you're going to get visible banding in the shadows. With 1000:1 you're essentially exposing a 90% white card at 10%-that's very, very dark, and you're only giving 400 levels to that top 90% range, excluding the noise that is present that far down in the luminance range. So your 11 f-stops might be a bit optimistic, I'd say we're probably at 10-10 1/2 f-stops depending on noise. Most likely 10 f-stops is the limit with tolerable noise levels.

[Juan P. Pertierra]

Joel,

I already did a test such as the one you are describing. I took a set of images at of the exact same scene with the same settings, while varying just the aperture. I took DV and RAW counterparts of each frame.

expert.cc.purdue.edu/~pertierr/LatitudeTest.jpg

You can also clearly see the artifacts in the redhat box. The only problem with that test was that I didn't adjust white balance, so the dynamic range is actually more than what you see there. White Balance works by limiting what the maximum value that comes out of each A/D is. So even though you can represent 0-4096 (12-bit) levels on each channel, you might only get an image with, say, 3980 as the maximum, washed out white on the RED channel for example.

Yes, and that makes me wonder....what color do I have to put in front of the camera such that a white balance adjustment will yield the absolute maximum range from all three A/D's? THe image color would look terrible, but it would allow the maximum resolution for working in post! It's only a little extra color resolution, but it sounds simple enough to find out.

Here's a PSD capture, uncorrected but with WB adjusted on the camera. Only thing I did was align the frames as much as I could.

http://expert.cc.purdue.edu/~pertierr/cap12_RAW.psd

[Juan P. Pertierra]

Oh yeah, and that is with the anamorphic, but it also has the overscan area since it is the full CCD, so you will actually get something like 1.80-1.85 aspect ratio.

Juan

[Joel Corkin]

Ah yes, Juan I forgot about that test you did. It's been a while. Thanks for reminding me of it. It looks really good aside from the white balance issue, which makes it difficult to compare dynamic range.

Interesting about the optimal white balance. I hope you figure it out. Sounds promising.

[Phil Rhodes]

Hi,

Something odd is going on there. Is it just that the blue channel is running into clip much faster than the others, or is there something odd we've all overlooked?

Phil

[John Gaspain]

Ive been away for about 6 months. Can anyone post a quick report on the progress? Last I heard was that Juan had captured a single frame from test jumpers on a chip. Any working models yet?

74 pages...sheesh!

[Nick Hiltgen]

John, I believe Juan is working on a web page that should answer all of your questions, now I'm not sure exactly when he will post it (he's trying to get the info accurate before he releases the website to the public) but hopefully it will be in the next couple days, I think september 1st would be a good goal...

In case that doesn't happen, the results are that juan has now captured a sequence of frames and almost completely elliminated the noise in them. He has re activated his white balance and the main thing we're all waiting and discussing right now is how much lattitude one can get with the camera. (is it plus 4 stops or plus 5, and how much did it have to begin with, 6?)

Hopefully that brings you up to speed enough until www.444yourdvx.com (or whatever it will be called) is up until then we just have to stick with www.waitforyourdvxtobe444d.com

[Mark Grgurev]

Okay Juan, now WAY WAY way back on like the first few pages, people thought the DVX could make a high def image. I remember you saying it can't, except maybe with some pixel shift. The DVX and the XL2 have horizontal pixel shift. How would I be able to use there pixel shift to make a higher res image using your mod?

[Thomas Smet]

The DVX already uses the pixel shift so you cannot use it again. Juan has shown us some images captured however that were blown up to 2k images and they look damn good. A program such as s-spline pro can make it even better. I however have used just Photoshop and even went up to a 4k image size with it still looking pretty damn good although a hair soft. Just having raw RGB pixels alone make the blowups look very good. It is sort of like if you have a 2 megapixel digital camera and it says you can get 8x10 photos from it. Well those 8x10 photos would be less than 150 dpi which isn't as good as getting a 4 x 6 at 300 dpi. However because digital cameras use RGB pixels the blowup looks much better that if you were to try and do it with a video frame. Juan's mod is basically allowing us to turn the DVX into a 1 megapixel digital camera that shoots 24 frames per second at perfect raw RGB.

So to answer your question you can blow up the footage to be 1920 x 1080 HD rez and it will look great except for a slight softness compared to actual HD footage. Some may actually like this softness because it makes the footage seem a little more film like instead of HD like.

[Juan P. Pertierra]

I second Thomas' statement.

Up-rezzing the full quality output of my device will simply give you the BEST possible enhancement without actually upgrading the sensors or other hardware. The only limitation is how good the up-rez algorithm is, and S-Spline is pretty darn good.

The full color sampling and high color precision transposes to the HD up-rezzed image. Remember that most of the HD footage we see is decimated, just like any footage that originates on a compressed format. Even the tape outputs on the high-end HD cinema cameras is decimated.

By starting with an undecimated SD image, you can obtain color comparable to a HD image that originates on HD chip(s). The downside, of course, is detail. S-Spline does a great job at enhancing detail but detail is simply limited by the resolution of the sensor.

My experience so far, however, is that S-Spline generates some awesome HD images when based on RAW images from my device, and it even introduces some subtle grain-like noise in the up-rezzed image which looks remarkably like film grain.

I've been looking without success for a location with a film printer, such that I can print a couple of frames to projection film and see what that looks like.

Juan

[Justin Burris]

Juan,

If you want a film-out, I would recommend putting the frames on a CD, taking the disc to a photo finishing place and asking them to make slides. I am pretty sure most places can do that.

Would that work for what you are trying to accomplish?

[Phil Rhodes]

Hi,

I think everyone's overlooking the fact that the DVX-100 is a 4:3 camera and under most circumstances you'd want a 16:9 image, or thereabouts, for your filmout or HD presentation. Not to say that this won't look a lot better than a DVX would normally in any case, but it's still a 40% cutoff. Might be time to reach for the mni35 with anamorphic primes.

Phil