5K 9K come on whats going on!?

[shannyla]

Genesis uses a 12 MP CCD sensor.
So does the Dalsa Origin.

There seems to be quite a bit of confusion between Bayer masking, RGB masking, single chip CMOS, single chip CCD and 3-chip Prism cameraas.

Either chip technology can be used with any of the color separation technologies; they are two entirely different things.

Having rechecked you're quite right about it being CCD, some info I came across had claimed it was CMOS and I'm guilty of not checking enough. And you are quite right about the confusion between colour separation and sensor technologies. I think the issue is that marketing has attached 3-chip to CCD and Bayer Pattern to CMOS in many people's minds, I think as an attack on CCD cameras.

That said, the Genesis is a very interesting application of CCD tech, and the pictures look incredible.

 

[shannyla]

At the native 4K size, yes, chroma resolution is only half, which makes RED's chroma resolution only 2K. Of course, that's still more color resolution than the 1080p cameras.

Anyway, if sampling color at half-res is so bad, why does Panavision specifically recommend 4:2:2 over 4:4:4 for post production? Even for special effects and green screen? They say the difference isn't visible for most applications. In the same way, Bayer pattern sensors simulate the human eye by having higher luma resolution.

For whatever reason (weak OLPF or a bug in thier magic), the Genesis output suffers some aliasing artifacts. 4K, on the other hand, lets you choose from a variety of software low pass filtration techniques as you resample to 1080, and isn't limited by the physics of birefringent filters like a native 1080p camera is. This nets an effective 15+% resolution gain without aliasing artifacts: 4K makes a better 1080p than a native 1080p camera.



My guess is that is the primary reason.

Totally agree on both counts, if anyone can actually see the difference between 422 and 444 on an image that has had little or no post capture processing then they have far better eyes than I do. Likewise the 1080 resolution is a chicken and egg question of what came first.

I'm not sure I'd agree with your statement about posting in 422 however. Going down that route would lose you the benefits of shooting in Panalog in the first place, and I've never heard anyone from Panavision UK suggest that route. That said, the Panavision Genesis Display Processor 444:422 convertor is a billable item...

 

[John Moores]

This is a common misconception. If the small pixels have proportionately smaller sensitivity (and FWC and noise), then when they are resampled to the same size as large pixels, the final image is the same.

But they don't tend to scale proportionately: they tend to scale *better*. For example, cheap little digicams have had gapless microlenses for a long time, which is only now starting to trickle down to DSLR cameras. For that and other reasons, the QE/area, FWC/area, and read noise per area on digicams is better than DSLR. For example, the LX3 has 50% higher QE/area than even the D3: 2.20 e-/16-bit raw level [4].

Eric Fossum, inventor of the CMOS APS, the technology behind almost all new DSLR cameras, coined [1] Agranov's Law: "More pixels improve image quality even in the presence of noise". This is based on a paper presented by G. Agranov at 2007 International Image Sensor Workshop, where it was demonstrated that pixels sizes between 5.6 and 1.7 microns all give the same low light performance, but improved performance with more light.

Eric also said that FWC tends to increase with smaller pixels [3]: "What we really want to know is storage capacity per unit area, that is, electrons per um^2. Generally, as technology advances to smaller dimensions, this number also increases. So, in your terms, smaller pixels have greater depth (per unit area) and saturate 'later in time'".

[1] Eric Fossum coining Agranov's Law: http://forums.dpreview.com/forums/read.asp?forum=1000&message=24828114

[2] G. Agranov's paper: http://www.imagesensors.org/Past Workshops/2007 Workshop/2007 Papers/079 Agranov et al.pdf

[3] Eric fossum http://forums.dpreview.com/forums/read.asp?forum=1000&message=30017021

[4] http://forums.dpreview.com/forums/read.asp?forum=1033&message=29866728

Thanks for info Daniel.
jm

 

[Daniel Browning]

I should first say that I've really enjoyed all the Genesis movies I've seen; the images it produces are very good.

I'm not sure I'd agree with your statement about posting in 422 however. Going down that route would lose you the benefits of shooting in Panalog in the first place, and I've never heard anyone from Panavision UK suggest that route.

The information came from John Galt, Senior VP of Panavision. It started when Ben Smithard, a London DOP, asked the CML (Cinematography Mailing List) if there was a noticable difference between 444 and 422. There was a wide variety of responses. Someone said that the color space was improved in 444, but that's wrong, and John Galt responded with a correction. After correcting the error, he went on to say that is it extremely difficult to demonstrate the difference between 444 and 422:

[...]
Although their is a theoretical improvement in color difference blue screen/green screen photography because of the equal full resolution of 4:4:4 RGB). It is extremely difficult to even demonstrate this with real images. Resolution in this context is bandwidth, or spatial resolution, not color.

John Galt
Senior Vice President
Panavision Advanced Digital Imaging

Several people responded to John with surprise and said that they get better matte extractions with 4:4:4. John said his tests showed otherwise:

My comment on color difference photography was not anecdotal, but based on extensive image composite testing done by myself and my engineering group.

The color difference compositing process for both film and electronic imaging was invented by Petro Vlahos, for which be has won two separate Academy wards, one for film and one for electronic compositing. Petro also founded the Ultimatte corporation. I think it is telling that Ultimatte's latest hardware compositing system no longer supports 4:4:4 but still enables beautiful composites of smoke, liquids and fire without artifacts.

Panavision has a film test which compares 4:4:4:, 4:2:2 and HDCAM image compositing. The subject matter includes liquids, flame, smoke and graduated color on a fine mesh cloth, all designed to exercise the process to its limits.

[...]
John Galt
Senior Vice President
Panavision Advanced Digital Imaging

John also said that he's not advocating that 444 be abandoned, but if it doesn't make a difference for even the most difficult green screen, I wonder when it would be useful.

Several months earlier, John also said:

I certainly was not trying to start an argument when I suggested that a Bayer pattern sensor might be considered to have either a 4:2:2 or a 4:2:0 sampling structure. I actually thought I was being charitable.

He went on to explain why he thought Bayer was even worse than 4:2:0. I disagree, particularly when the Bayer camera starts with a total number of samples that is quadruple that of the Genesis/F35.

The reason might be because Bayer sampling with advanced demosiac algorithms do very closely match the vision of the human eye, which has greater luma resolution than color. RED images demonstrate that, I think.

 

[shannyla]

Lots of stuff written by Daniel

I agree that video-gamma encoded images in 422 and 444 would be difficult to tell apart. However the best mode to use the Genesis in for dynamic range is Panalog, which although a fairly mild log encoding is still a log encoding and in my experience better results can be obtained in grading and processing with 444 log images, probably down to smoother transitions between color values and therefore less posterization. I will agree that its subtle. I'm not a compositor so matte extraction isn't my thing.

Perhaps 10-bit 422 and 10-bit 444 are comparable, I'd be lying if I said I'd ever even thought of doing a comparison between them, and John Galt is a clever guy.

He went on to explain why he thought Bayer was even worse than 4:2:0. I disagree, particularly when the Bayer camera starts with a total number of samples that is quadruple that of the Genesis/F35.

The reason might be because Bayer sampling with advanced demosiac algorithms do very closely match the vision of the human eye, which has greater luma resolution than color. RED images demonstrate that, I think.

That said, he does occasionally seem to have some strange ideas. I once had a conversation with him about the SSR1 recorder, where I expressed my surprise that you couldn't access the digital media directly as files, that you have to play out from the solid state recorder to an SR tape. He couldn't see my problem with that.

Does a Red have quadruple the number of samples that a Genesis does? At 12 megapixels I'd say they were roughly the same. Remember that the Genesis is a true 1920x1080xR, 1920x1080G, 1920x1080B. Effectively the sensor is 1920x3024... My opinion with regards Bayer sampling is that the quality of the decoder has a disproportionate effect on image quality. This is blatently obvious in the digital stills world, and there is no reason why moving imagery should be any different in this respect.

 

[Daniel Browning]

Does a Red have quadruple the number of samples that a Genesis does? At 12 megapixels I'd say they were roughly the same. Remember that the Genesis is a true 1920x1080xR, 1920x1080G, 1920x1080B. Effectively the sensor is 1920x3024...

You're right. 4K is only quadruple the spatial samples, but if color and spatial samples were weighted equally, RED only has 35% more (4096x2304).

My opinion with regards Bayer sampling is that the quality of the decoder has a disproportionate effect on image quality.

Agreed. If you need real-time 1080p output (e.g. for a tape recorder), with meager computer hardware, it's very hard to get the quality high enough with Bayer, whereas RGB stripe requires zero computer power.

On the other hand, if you can record the raw Bayer information and spend lots of computer time and heavy computer power (more than you can fit inside a camera) on the demosiac, then the quality is outstanding.

 

[Graeme Nattress]

RGB stripe though "should" require processing to spatially co-locate the samples. If you're resigned to the fact that you've got a single sensor with a colour filter array, the proof is in the pudding that arranging those samples as a Bayer pattern, gives a higher image quality than the RGB stripe method.

Say you take that 4096x2304 and do a Bayer pattern on it. Measured chroma resolution on pure colours (you'll get much better results on any non-pure colour, but let's take worse case here) is 2048x1152. Luma you'll get about ~78% of 4096, which is 3200x1800.

Arrange those 4096x2304 as a RGB stripe, you'll measure luma and chroma the same as 1365x2304. Actually you'll measure less as you'll need an OLPF in there. But let's give them the benefit of the doubt here. We need to make the pixels non-square to get them to produce a 16:9 image with equal resolution, and that'd give 2352x1323, but in reality less. The Genesis / F35 arrangement will get lower resolution still as it's using a 6 pixel macropixel, giving 1672x940 from the starting number of 4096x2304 pixels.

Given the way the eye sees things, Bayer patterns put much more relevant resolution to that eye. Viewing the 4k movies in the RED cinema, the resolution advantage is pretty obvious too.

Graeme

 

[oldphart]

RGB stripe though "should" require processing to spatially co-locate the samples. If you're resigned to the fact that you've got a single sensor with a colour filter array, the proof is in the pudding that arranging those samples as a Bayer pattern, gives a higher image quality than the RGB stripe method.
...

Graeme

If we disregard real-life practicalities, would not the stripe pattern have an advantage if the image is displayed on a device with the same pattern at the same resolution?

 

[Jon Schellenger]

I just came across this post...

Trust me. There is a reason for this - even the 28K. Soon everyone will know.

For me, count me in. I am very much into something at that resolution.

And I disagree with the quote that the RED is not the same film like quality as Dalsa, or others. Just look at the trailer for Knowing. Case closed!

 

[J. Eric Camp]

Both Genesis and the D21 are CMOS. D21 works exactly the same way as Red, Genesis uses a 12mp row array sensor which is arguably better than a bayer pattern sensor in terms of color accuracy. These aren't 3CCD HD cameras which is what you seem to be suggesting, nor are they "older" HD tech in any way, unless you consider a S35-sized single CMOS sensor to be in some way obsolete. You could maybe argue about recording onto tape, but SR is reliable, known, proven in production, easy to post and easy to archive. The only downside to HDCAM-SR is the cost of the decks but in every other way it's a wonder format if you've been around this business for any length of time.



You're misunderstanding the purpose of the Genesis (and by definition the Sony F35, for they are the same...) resolution. It seems to me that Panavision decided, with good reason in my opinion, that for moviemaking colour resolution was more important than spatial resolution. So for each pixel there is a dedicated red, green and blue sensor. In a bayer pattern sensor like the Red and the D21, 50% of the photosites sample green, 25% each for red and blue, and the quality of the raw decoding software determines your color accuracy.

The other reason that the Genesis is a 1080 camera is because thats the format of the SR deck that is used to record the signal.

Don't knock the Genesis until you've tried it, of all the digital cinema film replacement cameras it's the one that I would be most comfortable, in terms of the things that professional filmmakers enjoy such as say "will it work when I turn it on..?", shooting a production with. And we've shot productions with literally all of them, starting with the Viper in 2002.

Number one: I said they are all good cameras.

Number two: I said the genesis is built very well.

I very politely discussed how I view it. So don't misconstrue my thoughts and comments for knocking or bashing the Genesis. It takes pretty pictures.

Now then... I know how a bayer pattern works and I also know how the Genesis sensor which is NOT A CMOS SENSOR works. I Never suggested it was a 3chip CCD, but it is infact a CCD. It works through a combination of RGB strip and pixel binning. The sensor is exactly 5760x2160.
1920 x 3 (RBG) = 5760
1080 x 2 (pixel binning) = 2160.

Six pixels on the sensor for every pixel in the final product.

I said, all systems have their benefits, all different ways of skinning the cat. The camera works, and I am not knocking it. Nor the D21 which is a CMOS sensor.

The reality is, all cameras can be used well if the jack ass behind it puts the effort in.

 

[Graeme Nattress]

If we disregard real-life practicalities, would not the stripe pattern have an advantage if the image is displayed on a device with the same pattern at the same resolution?

It's called a Trinitron Tube TV.... Sit far enough back and there you are....

But you can see, from the above, that you just don't get the resolution...

Graeme

 

[oldphart]

It's called a Trinitron Tube TV.... Sit far enough back and there you are....

But you can see, from the above, that you just don't get the resolution...

Graeme

I replaced my Iiyaku and Sony Trinitron monitors with Eizo LCD. Still has the same stripe pattern, just 2k wide. Or 6k if you count colour stripes.

Still, I trust your cat-skinning by the results you obtain. Guess you can skin the same cat up to nine times, though, so I expect you to eventually come up with an even better way. Two different green filters? One Y an one C instead of the two greens? Suppose much depends on possible filter quality. It is one thing to play around with software modelling, a totally different thing to realize it physically.

 

[Graeme Nattress]

I'm sure that in the future there will be "better ways" but until then....

Graeme

 

[Radoslav Karapetkov]

Mr. Nattress, do you think the new black silicon thing could improve those issues you said about the FoveonX3 chips?

About silicon being a poor color filter, etc.?

And, do you think that a similar kind of layered display is possible?

I mean, a display that has three layers of pixels, each behind the other. You know, to imitate the way film produces colors in projection.

Would it be practical?

 

[Graeme Nattress]

I don't think anyone knows enough about black silicon yet to know how it would help, but as it's a surface treatment, and the Foveon uses the depth of the silicon to work, I can't see it helping it's colourimetry one little bit.

For projection, I don't see anything wrong in the colourimetry of current high end projection.

Graeme