4:4:4 10bit single CMOS HD project

[Obin Olson]

David could your codec support 1080P editing with Premiere Pro? in realtime? with 10-12bit?

[Juan M. M. Fiebelkorn]

Obin, could you show me what is the "Smear effect"????

[Obin Olson]

I will get you a snapshot when I get the camera from work

[David Newman]

<<<-- Originally posted by Obin Olson : David could your codec support 1080P editing with Premiere Pro? in realtime? with 10-12bit? -->>>

Yes, it basically does that now with the product Prospect HD, 1920x1080p24 10bit real-time, but you probably know that by now. I've been working on the smallish modifications to support bayer direct encoding, this would be an extension to Prospect's features. We have already discussed that the turn-key Prospect HD system from BOXX is too pricy for this market, so we have been thinking of ways to introduce a lower cost product that doesn't hurt the higher-end market -- we are still working on this marketing issue (that part always seems harder than completing the technology.) There are some cool ideas in the works, nothing I can mention yet.

[Les Dit]

I think a little explanation might be in order:

Imagine if I make an image with a paint program, and I use only the following pixel values: 0, 1365, 2730and 4096.
Only 4 distinct values. Very 'posterized' looking.
Ben, do you see how this is not a 12 bit image?
It's a 2 bit image. It only has 4 code values. I could save it as a high bit depth TIFF, but that does not make it a 12 bit image just because it has values of 4096 or whatever.
It has lots and lots of missing code values.
Now, imagine a 10 bit image that has a thousand possible values, but spread them out to still reach 4096. The image will look kinda normal, but it's still a 10 bit image.
A true 12 bit image has no missing values if there is a shaded ramp in the image. You can verify with a histogram, of just look at pixel values with a viewer other than photoshop. All values should be there, sprinkled with a dose of noise, of course!
( I'm talking about a RAW monochrome image with no demosaic applied, BTW )

End of "basic Bit depth for dummies" 101..... ;-)

-Les


<<<-- Originally posted by Ben Syverson : I have no idea. All I know is that the data in this file goes from 0 - 4096. -->>>

[Ben Syverson]

@Les: End of "basic Bit depth for dummies" 101

I don't need the lecture. I think I understand bit depth as well as anyone on this list. I'm the one who figured out it was 12bit data in a 16bit file fer chrissakes.

Begin Advanced Bit Depth 159, lecturer: Ben Syverson.
If you open the image in Photoshop as a 16-bit RAW file, you'll see it's dark -- if you do a levels with an input white of 16, you'll expand 0-4096 to the full 16bit range. I'm not sure if PS uses full 16bit (65535) or After Effects-style 15bit+1 (32768), but either way you're not losing any data.

Now do a levels call like input black: 100, input white: 116. This is to see if the image is 10 bit or 12 bit. If it's 10bit, the histogram will show banding. If it's 10bit, the range between 100 & 116 should fill up the histogram. In fact, it does. The data is 12 bit. Don't ask me how it got there, I'm just telling you what's in the file.

And why wouldn't it be 12 bit data if the max value is 4096? Why on earth would they output 10 bit data to a 16 bit file but multiply it by four? That makes no freaking sense. If you were going to multiply the data anyway, why not multiply it to fill the full range of 16bit? Why stop at 4096?

The reality is that it's simply 12 bit data padded out to 16 bits for ease of use. Save the condescension for someone else.

- ben

[Rai Orz]

Ben, it looks like LES know what he said. And i know, here the people who make the post, also use histograms to check bit deeps. Send LES the test frames and we will see.

[Les Dit]

Sorry Ben,
So if you saw 12 real bits, no missing code values in the RAW image, there are two possibilities:

The camera is making 12 bits ( perhaps 2 bits of injected fake noise! ) and SI didn't know ? !!

Or:
The image isn't really RAW and was tampered with.

Where is that Steve guy ?


Bring it on!

Send it to my gmail account. Gotta use up that free gig they gave me ;)
lesdit gmail com ( fill in missing stuff in address )

-Les

[Ben Syverson]

Les is right -- after looking at it more closely (Photoshop's histograms do lie), this is in fact 10 bit data, multiplied by 4, and stored in a 16-bit file.

That's got to be the worst idea I've heard all month. I think I just couldn't believe that was the case. That and I was calculating for 65535 when I'm pretty sure PS maxes out at 32768.

If you open up the image in After Effects (where we know for a fact that the max value in 16bit is 32768) and set the input black to 256 and the output black to 512 you'll see banding in the histogram (you have to apply another Levels in order to get the histogram). Those 256 values would fill the histogram if we were dealing with 12-bit footage. Instead, there's a spike for every other value -- it's half what 12 bit should be. Since it's already been cut in half by AE, the histogram shows 1/4th the data of 12bit. In other words, it's 10bit.

Utterly ridiculous...

[Ben Syverson]

In other news, I did shoot some tests tonight. When they say the IBIS-5 is low-sensitivity, they mean it. It's similar to shooting with 100 ASA film. If you're not outdoors, you better light the heck out of your subject.

I'm still getting used to the controls, so these images clip at black. The thing is, the camera outputs downsampled 8bit from 10bits. We all know how hard it is to work with 8bit files, so with this setup my strategy is to do as much of the image/color adjustments as possible on the camera. The software basically controls a 10->8bit look-up-table, so you can do things like gain and gamma in 10-bit before it gets knocked down to 8. Anyway, unlike 10 or 12 bit, you're not creating a "digital negative" that you can tweak to your heart's content -- the image you capture is pretty close to being the final image.

At 40mhz (the top speed), the image gets kind of unstable. I think it's an issue with the USB2 bus on my laptop, because I tried it on a CardBus USB2 card, and it was even worse. But 24mhz is pretty much perfect for my use. At 1280x720, you get almost exactly 24fps. Since 1280 * 720 is so close to 1,000,000, you can pretty much look at the mhz, and that's what your frame rate will be. So at 40mhz, it hovers around 39fps.

Direct-to-disk recording is way slow right now (4fps), possibly because I'm using the 7200rpm drive through USB2. Hopefully that will improve once the drive is installed as the second internal (the HD bay is coming on Mon/Tues). But the software has a built-in Record-to-ram function, and it works great. I'll be returning the 256mb chip I just bought and buying two 512 chips instead (the max this laptop can do is 1gb). Even if that's my only option, figure 700 megs free out of 1024, at 21megs/sec, that comes to around 30 seconds. That's workable.

And my cute Computar lens is not the greatest piece of glass ever to grace the film plane. It's awesome at around F4 -- anything under that, and it's diffused. It's weird -- it doesn't go soft, it just starts to look like you're shooting with a ProMist, and it changes with the focus. The good news is that 8mm with a 2/3" sensor is really nice and wide.

The real champ in my toolbox right now is the ancient Angenieux 25mm f0.95. At f0.95, it's over a full stop faster than the f1.4 Computar. In other words, it lets in over twice as much light. When dealing with this sensor, that's a life-saver. Also, it's far better at every aperture than the Computar. It hits a sweet spot around f2, but even wide open it's gorgeous.

Case in point:
phone.png -- Angenieux 25mm

Here's a close-up test:
face.png -- Angenieux 25mm

For comparison, here's an image taken with the Computar from the same distance as the last shot. This gives you an idea how wide it is in comparison.
head+shoulders.png -- Computar 8mm

You can see how washed out the colors look. I kind of like it -- it's like a bleach bypass effect. The controls are a little picky, but this camera is definitely usable. I have to plan something to shoot on it now.

Tomorrow I'll head outside and take some daytime shots around the neighborhood. We'll see if the colors stay washed out in the daylight.

- ben

[Rai Orz]

The adventure begins, please help me...

I will order a camera set next monday. Also some hardware and maybe software.

But i am not shure about the camera. Obin and some others use the SI-1300 but now Obin wrote "... I don't think any of you could shoot professional stuff with this chip..."

What´s going on? Maybe he had a optical (lenses) problem, maybe with UV-light. Or is the camera unusably?

Before i order a camera - set i need sample uncompressed video pictures (1280x720, 24 or 48fps, 10bit) as RAW File and maybe 24frames/second as TIFF or TGA, from all cameras from any of you here. Is this possible? We know the different between good or bad lenses, but we dont know the data details comming out and can save to Disk.

We need only 24fps/ what is the thing with 48fps?

The next question: My software people said: It will be better to use a matrox grabber-card with 32MB on board (because lost frames). Matrox also have grabber software tools and with matrox germany we can work together to develop a own software.

I am not the man for software, but i and my company made opto mechanical parts and solutions so we can make (and have) all things to build a prof. solution. Everyone who can help me in software and storage problems can have parts from us, or we can work together on a different way. Also if the first movie shoot with this kind of camera, dvinfo and all this people can be mentioned.

Till monday i must have all details.

Thanks to all

[Eric Gorski]

those stills look very promising ben. i messed around with them in after effects alittle and they can actually be color corrected quite a bit... if you post some video, could you please shoot something with some fast pan or something, to get a sense of strobing? thanks.

good work.

[Obin Olson]

ya Ben cheap lens! Computar! I have one and it's the same gets all fuzzy at wide open..weird huh?

I really need to check with that chip of mine and do some tests with hot spots in the frame. It SUCKS when you get smear all across your image!!!


Rai:
I will try and post my shot from last night on the dvx100 so you can see what sorta of shot is bad with this chip..just an fyi

[Jason Rodriguez]

Ben,

Does their camera output RAW Bayer files? It seems like their software isn't so good at the demosaicing stuff, and while your software does remove a good portion of the artifacts, it can't seem to remove the aliasing and color moiré.

BTW, that RAM recorder sounds like a really good idea. You can buy 2GB DIMM's now, and maybe with some smart hard-drive buffering you can get a little more time out of the chips. But I would think those small ITX motherboards could take a 2GB DIMM, at least one, and maybe two, for a total of 4GB? If you could get 4GB on a board, that would be about 1 minute at 24fps 1920x1080 on the Altasens.

[Obin Olson]

I think RAM recorders are for highspeed filming...in short bursts for slow motion stuff