Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Why Bayer

The point I'm making is that the result != spec in terms of measured visible resolution. Put a Kell discussion in off topic if you want!

Graeme
 
Graeme Nattress said:
The point I'm making is that the result != spec in terms of measured visible resolution. Put a Kell discussion in off topic if you want!

Graeme
Click to expand...

Sorry, did not mean to offend you. BTW, I must say that your first post on this thread and on the Cinematography list was excellent. Really enjoyed reading it and was quite informative. Thanks.
 
Graeme Nattress said:
Say you have 12mp of photosites. What is the best way to arrange them?

3x2.66k chips (the prism + three chip approach)
12mp = 4.61k bayer pattern sensor (what we do, near enough)
12mp RGB stripe, 4.61k (Genesis approach)
Click to expand...

Sorry... but I'm a little confused by the 3 x 2.66k in that list? Why would it be 2.66k and not 3 x 4.61k?
 
Because the problem was posed by "Say you have 12mp of photosites...". If you had 3x4.61k that would be 36 Mpixels total, not 12. 3-CCD cameras usually do call out pixels per chip rather than total, but if cost is the constraint then it's total silicon die area that matters, not just pixels per sensor, plus the prism limits your overall size anyway.
 
jbeale said:
Because the problem was posed by "Say you have 12mp of photosites...". If you had 3x4.61k that would be 36 Mpixels total, not 12.
Click to expand...

To me it's not, it's 12Mp per sensor so you need 3 sensors to collect full color data.... meaning you only measure the res. of one sensor... while you're just using 3 to collect full color information. How does a prism limit size? Or do you mean size of the entire camera?
 
Andrew McCarrick said:
To me it's not, it's 12Mp per sensor so you need 3 sensors to collect full color data.... meaning you only measure the res. of one sensor... while you're just using 3 to collect full color information. How does a prism limit size?
Click to expand...

You can choose to count pixels that way, but if you do that it would be as appropriate to count the R,G,B pixels in a Bayer array as separate sensors too, that just happen to be located on the same physical substrate.

In theory you can make a prism any size but in practice there just are no 3-CCD type prisms made larger than 2/3". I don't know if it's cost or something more fundamental.
 
jbeale said:
You can choose to count pixels that way, but if you do that it would be as appropriate to count the R,G,B pixels in a Bayer array as separate sensors too, that just happen to be located on the same physical substrate.
Click to expand...

Sure, that makes sense.
jbeale said:
there just are no 3-CCD type prisms made larger than 2/3".
Click to expand...


For now anyway.... :whistling:
 
Joofa said:
Sorry, did not mean to offend you. BTW, I must say that your first post on this thread and on the Cinematography list was excellent. Really enjoyed reading it and was quite informative. Thanks.
Click to expand...

No offense - just trying to keep it on topic. I always enjoy discussing this stuff, and Kell gets quite esoteric and subtle.

Graeme
 
There's (to me) two things that confuse me, as a very definite non-engineer.

1. The bayer issue. Is luma full res?

2. There's really so many factors to the effective resolution, where glass and exposure are the most important, given that you have a sensor like REDs. What are the real gains with EPIC, except for less compression (which to me is the main selling point... or at least one of them :) )

If this is OT, just ignore...
 
Gunleik, the issue is not that bayer doesn't give full rez, but that the only way to get full rez also gets you aliasing, which is bad! My whole argument is about if you only have so many photosites, what's the best way to spread them out. In reality, in real camera design, number of photosites is a real thing - it effects memory, buffers, compression, speed getting data off sensor etc. etc.

I hope to eventually produce images to illustrate my argument, but time does not currently permit.

Graeme
 
Graeme, this was not at all a complaint! :)

I'm so stoked by what I get, that I cannot really see what I can complain for (except bad exposures, which is by no way REDs fault :) )

I'm living in a world were 100% (not close to, but 100%) of finishes, even for film is done @ 2k @ best.

My point 2 does (to me) really affect imaging a lot more. RED is capable of of resolving most things I've been throwing @ it so far. People seem too (IMHO) hung up on "is 4k really 4k" and too little hung up in exposure/lens issues, where we've eventually come to with RED.

The lesser compression in EPIC is still a valid thing, though. Less compression is not only bad, not for recording formats under non-optimum conditions.

Cheers.

gettin' meself another beer now.
No whiskey here tonight...
 
why 5K

why 5K

darkkelt said:
I know this is probably slightly skewed but if the effective resolution is 3.2k because of the way bayer and other bits are done how do we gain anything going to 5k on epic ?

or is their some extra pixel magic dust being done and we'll find out later ?
Click to expand...

I'm algorithmically challenged, but it seems to me this was a smart choice in that you end up with actually 4K of resolution, which is good assuming we get a true 4K output/display pipeline. Yet by not going much more than 5K you minimize the effect of reducing pixel size that has too happen as you increase the number of pixels on the same size chip. The loss in sensitivity to reduced size (since it is not a huge percentage of the original 4K chip) may be offset by improvements in the sensitivity and DR of the next generation chip.

How's my thinking?
 
Graeme Nattress said:
Ronx, you're getting sampling theory wrong. A 4k image can accurately hold < 4k detail, not <8k detail.

Graeme
Click to expand...

I didn`t say it could. I said an 8k bayer is needed to create something close to a true 4:4:4 RGB at 4k. To get a 4k resolution of green, you need twice as many green pixels than a beyer array has, and for red/blue, four times as many than a beyer has. So you`d need an 8k bayer to remove most of the guess work done by the demosaic. Such an 8k CFA would have 4k worth of green pixels and 2k of Red and Green = 8k of pixels. A 4k RGB 4:4:4 image has 4k of green, 4k of red and ,4k of blue=12k of pixels.
 
To get a measured 4k luma resolution we need only 5k though. An 8k bayer sensor would give you luma resolution out to about 6.4k, and if you did that with a smaller pixel pitch, you may even get away with a weaker OLPF as of the the lesser ability of the lens to resolve out there.

Graeme
 
So at what point does a lens become a good OLPF rather than a crappy lens?

So at what point does a lens become a good OLPF rather than a crappy lens?

Graeme Nattress said:
To get a measured 4k luma resolution we need only 5k though. An 8k bayer sensor would give you luma resolution out to about 6.4k, and if you did that with a smaller pixel pitch, you may even get away with a weaker OLPF as of the the lesser ability of the lens to resolve out there.

Graeme
Click to expand...

So at what point does a lens become a good OLPF rather than a crappy lens? Older cine lens? Newer SLR lens (doesn't seem like it)? ??
 
Who knows? I've not really profiled lenses to see....

Graeme
 
To get a measured 4k luma resolution we need only 5k though. An 8k bayer sensor would give you luma resolution out to about 6.4k, and if you did that with a smaller pixel pitch, you may even get away with a weaker OLPF as of the the lesser ability of the lens to resolve out there.
Click to expand...


For luma and practical perceptual purposes yes, I wasn`t being accurate, but for theoretical 4k "colour", I still say 8k (which would make green pixel resolution 6k), then average 2x2 pixel blocks->4k (which would anti-aias image and reduce need for such strong LP filtering). I hope you don`t feel you`re defending Red1, cos I`m certainly not attacking it, just playing with the numbers (theoretically).

While you`re around Graeme, I don`t suppose you have:
Any idea if the current sensor will ever reach 60fps@4k. I need to know either that, or the estimated/ballpark price of the proposed sensor upgrade that will produce 60fps@4k, before I can finalise my order (before april 20).
ie, perhaps in the form of a non-binding guesstimate such as, "probably less than $1k/2k" etc would suffice.

I will be buying a
 
Yes, if you want 4k chroma on a Bayer pattern sensor, you'd need an 8k sensor at which point you'd also get 6.4k luma. I'd never demosaic by averaging block as that is a poor downsample filter and will create aliases. You can't anti-alias an image in software to make up for lesser OLPF - as the aliases occur at the point of sampling (ie when the light hits the sensor) and downsampling is still a sampling process, so all you can do is avoid creating more aliases, but the ones that got into the system are aliases you're stuck with.

The current sensor does do 60p at 4k, but we don't have the processing power / bandwidth in the rest of the circuitry in the camera to deal with 4k at 60p, but we can do 3k at 60p, hence we put that mode in.

Graeme
 
I think it should be possible to derive the max luma resolution possible considering the size of photosite and sampling grid constrains. Top of my head I would think it would reduce to a Bessel function of certain kind. However, I could be wrong. I need to have paper/pencil/and access to a library/etc. to do this calculation -- too tired to do any of it right now.
 
Graeme Nattress said:
Yes, if you want 4k chroma on a Bayer pattern sensor, you'd need an 8k sensor at which point you'd also get 6.4k luma. I'd never demosaic by averaging block as that is a poor downsample filter and will create aliases. You can't anti-alias an image in software to make up for lesser OLPF - as the aliases occur at the point of sampling (ie when the light hits the sensor) and downsampling is still a sampling process, so all you can do is avoid creating more aliases, but the ones that got into the system are aliases you're stuck with.

The current sensor does do 60p at 4k, but we don't have the processing power / bandwidth in the rest of the circuitry in the camera to deal with 4k at 60p, but we can do 3k at 60p, hence we put that mode in.

Graeme
Click to expand...

Yes, but you`d anti-alias compression artifacts, and the LPF would only have to filter frequencies twice that of the LPF used by a 4k beyer, as there`d be no need to LP frequencies the sensor can resolve?

I think you just told me Red1 won`t ever do 60fps@4k? Oh. I thought 3k was implemented "just because we can" and not a replacement for 30fps+ @4k. So, I need to wait till the new sensor is out and faster DSP processor etc, before I buy. What would be you`re personal best guess for the price jump (ballpark etc) of a Red 1 (v2) when the new sensor is out?
 
You must log in or register to reply here.
Back
Top