4:4:4 12-bit Uncompressed DVX100

[Juan P. Pertierra]

As for the design right now, all it cares is that the drive is an external Firewire 800 drive and it is fast enough to handle the bandwidth. It has no means to power the drive from the camera, because all Firewire 800 drives have different power supplies, so it would be impossible to predict what kind of drive the user will buy to use with the system. I would have to limit what brand/model of drive you can use with the device in order to provide power for it, and I don't think that is a good idea.

One idea is to provide batteries for different drive systems. They all have different plugs and voltages and such, but in the end it is not terribly difficult to put a battery system together for one of these drives.

We could even provide a small adapter, one end which plugs into the power of whatever drive the user has, and the other end plugs into one or more generic batteries such as RC Car batteries. I did this once for a portable radio application and it worked very nicely, specially for demanding applications. The batteries and drive can then be carried together in a shoulder bag, with only the firewire cable exposed.

Juan

[Chris Sorensen]

Since the NTSC pixel yield is 773 X 494, am I correct in assuming that without using the ana, or in-camera squeeze or letterbox, you can get a 773 X 435 16X9 image by cropping in post.

Also, since the resolution is higher than DV, is the CCD picking up more than the viewfinder/monitor would show, or does it pick up the same image and downsample it to DV? And if it's a larger picture area than a monitor would show, how difficult does that make framing shots?

[Brett Erskine]

Close enough:

773 divided by 1.78 (otherwise known as 16 X 9) = 434.27




Yes your monitor wouldnt show the entire picture that the CCD is capturing. I would think a overscan monitor would work fine but then again thoughs may not farther then showing you the edge of a NTSC standard frame (720 X480). Anyone?

[Chris Rubin]

Looking at example frames Juan has posted, it seems the 773 X 494 resolution does not represent correct aspect ratio: the images coming from the ccd have a sort of backwards effect of an anamorphic adapter. In order to correct it, the image has to be resized to 720 X 540 (or any othe size fitting the 4x3 aspect ratio.

peace,
Chris

edit:

I'm speaking in terms of square pixels here... on an NTSC monitor with 0.9 pixel ap, the image should appear 4x3. I do not have one (PAL country), so I'm not sure.

[Thomas Smet]

the 773x494 image still uses non-square pixels so you actually only have 687 x 494 raw square pixels. 16x9 HD uses square pixels so a 853 x 480 HD frame is square pixels. This is why cameras such as the XL2 have 960 pixels for 16x9. 960 equals the non-square size for 16x9. Once you convert the 960 to square pixels you get 853. yes you could use every pixel as is in 773x494 but then your actors would hate you because they would look fat.

So as you see we are only gaining an extra 47 square horizontal pixels in the image from the raw capture. 687 compared to 640.

Remember you can't really compare 853 to 720 because they are different pixel formats. 853x480 has to be compared to 640x480 or 720x540 if you want to blow up your vertical resolution with interpolation.

[Juan P. Pertierra]

Ok, I need everyone's help here:

In the latest tests i've been doing, I am getting a properly exposed DV image of a well-lit scene at F16, with some minor washout areas, but I am getting the same frame properly exposed with NO washout at F6.8 for RAW.

I highly doubt that I am getting 7 F stops of added latitude, so perhaps this method is not accurate? Any suggestions? Anyone?


About the frame size, on the DVX the CCD"s yield 773x494, in NTSC aspect ratio. But the lens itself stretches the image a but horizontally to take advantage of the wide CCD. So, after applying a 0.9 ratio for NTSC, the image needs to be compressed horizontally until a 4x3 aspect ratio is reached, plus the overscan area.

Juan

[Kevin Good]

That's not seven stops the way I count.

The best would be if you could post another example where you put side-by-side of DV-out vs. RAW-out at a bunch of different exposures on the same scene,

You did one of those before in the thread, but before the white balance was working, so it gets pretty goofy.

Try to get something with insane dynamic range. An unlit room looking out towards the window at daytime or something. Like:

http://www.theasc.com/magazine/image...cts/Prod3D.jpg

Although I can understand if you don't wanna bother with the model and the macbeth chart. :)

[Justin Burris]

Yeah, f6.8-f16 is three-and-a-half stops.

I concur with Kevin. That sort of test would tell us much more. If you happened to have a lightmeter on hand to tell us what the different areas were reading, then that would be even more useful.

[Juan P. Pertierra]

Ok...well it just happens that I found a local camera shop, so I'm just gonna invest in a light meter and solve this once for all.

Although 3 1/2 stops sounds about right.

Juan

[Justin Burris]

Cool. This type of notation makes it very clear:

http://www.theasc.com/magazine/image...cts/Prod3A.jpg

[Justin Burris]

You probably already know this, but just in case: if you are going to get a lightmeter, a Spotmeter is the way to go for these tests.

[Phil Rhodes]

Hi,

Just out of interest, what filesystem are you using on the hard drive? I ask this because I don't know what hardware you have in the device and if it isn't a full microprocessor system, implementing different filesystems is presumably quite a lot of work. What's ideal for a WinNT user (NTFS) won't be ideal for a Mac user. I believe both can read FAT32 formatted volumes, and that filesystem has been implemented in embedded systems before. You aren't just doing raw sectors?

Phil

the difference between f6.8 and f16 is 2.5 stops (5.6, 8, 11, 16). of course that's assuming that the dvx100's fstop ratios are truly at a doubling ratio... i have no idea if video cameras are expected to be accurate in that realm.

and the difference in latitude (between the raw and dv25 frames) based on fstop may be misleading. more than likely, there is indeed extra lattitude beyond the total black and pure white of the dv footage being laid to tape. but the chances that it exceeds 10 stops is very low.

i would assume that the raw frames (that appear darker) are being sent to the dsp (or whatever it's called) and are having the almost blacks and almost whites clipped and the gamma adjusted (to make the midpoint brighter). the clipping of toe and shoulder are probably a built-in quality control issue to compensate for slight variations in the ccds from camera to camera. and the gamma correction/midpoint brightening is standard proceedure for dealing with linear images.

so basically, i'm guessing that if you take a raw frame and adjust the white/black points and adjust the gamma, you could result with a frame that has similar "exposure" and shadow noise as the equivelent dv frame.

if you think of a semi-equivelent in shooting film, exposing at f6.8 would be overexposing, thus minimizing shadow noise, and increasing color saturation particularly in the shadows, but at the potential cost of blowing out highlights. exposing at f16 (based on the fact that the camera considers this a "correct" exposure) would be like a "normal" safe exposure that is balancing between maintaining highlight and shadow detail. just keep in mind that the raw frames are (i'm 90% certain on this) linear, so they will by default look dark. raw frames are not intended to be viewed by human eyes without gamma or color correction, so exposing your footage based on how the "straight out of the box" raw footage looks is probably not a good idea.

i would liken the raw frames to a film negative... it contains more image information than you will probably ultimately be using in your final piece, so exposing for the freedom of options in post is usually the best choice.

i was a cinematography emphasis in college, am fluent in the zone system, have shot negative, reversal, and dv professionally and for personal work and also have done post color grading professionally. i'm more than happy to try and help settle the latitude issue and/or the raw-to-dv exposure discrepency. if you'd like my input juan, just send me a raw frame with considerable luminance variation (the asc image kevin posted is a great example), along with the dv frame equivelent.

i'm not at all bashing juan's work or his mod -- i'm probably more excited about this than anyone else. i'm just trying to help everyone with whatever knowledge or experience i have. and regardless of the mod's true latitude, i can say that based on tests i did with the old frames juan posted, the raw frames give us incredible flexibility in color manipulation and the ability to adjust the luminance in the shadows and highlights with minimal side effects (banding, noise, etc.), especially when doing color correction in floating point or 16-bit software. in my opinion, that's way more valuable than having 10+ stops of latitude.

and juan, if you are going to do a scientific latitude/exposure test with a spot meter, i would highly suggest you ask an experienced professional photographer, preferably fluent in the zone system, to assist you. though it's perfectly possible for anyone with a spot meter to do it correctly, there's a lot of ways to accidentally taint the results.

hope this helps,
jaan

[Mark Grgurev]

How's the website going, Juan?

[Juan P. Pertierra]

The code is done, I am ironing out the content...for example I want to get an accurate measure of how much additional latitude I am getting. i do not want to make false claims on the site, so I'm trying to make sure everything is as acurate as possible.

Also, i'd like to take the chance to thank everyone for their continued interest, support and patience. I'm trying to do this as fast as I can, but it's just me :)

Somebody once said: fast, good, cheap...pick two. Hopefully this will give good(great) quality for a cheap price, but it takes a while.

Juan