NEW YORK + EPIC

[Roberto Lequeux]

Jean, your dark B&W keeps catching my eye. I really like it.

Not sure what the size of the image has to do with dynamic range or bit depth...

It is my understanding that when you scale down, say by 50%, you also double your rounding error accuracy. Each resulting pixel is pushed through the grade's math as two distinct pixels, and the resulting values are then averaged out into one. Though I understand enough about bit-depth to know that each additional bit per value has a much larger effect than halving rounding errors when it comes to grading. I mentioned it because it does help, even if only a little, so heck I'll take it! And because it really matters a whole lot in other ways like cropping in, and more importantly sharp images without the need for sharpening.

So what did the bit-depth on Epic end up being? Is it 16-bit?

Thanks. Jarred's Lighting helps a lot. :smile5:

Yes. Thanks again Jarred for letting us get a taste of grading Epic with such nice images. Very generous of you.

 

[jake blackstone]

J
It is my understanding that when you scale down, say by 50%, you also double your rounding error accuracy. Each resulting pixel is pushed through the grade's math as two distinct pixels, and the resulting values are then averaged out into one. Though I understand enough about bit-depth to know that each additional bit per value has a much larger effect than halving rounding errors when it comes to grading. I mentioned it because it does help, even if only a little, so heck I'll take it! And because it really matters a whole lot in other ways like cropping in, and more importantly sharp images without the need for sharpening.

So what did the bit-depth on Epic end up being? Is it 16-bit?

Yes, Epic produces 16 bit RGB images.
I wonder, if with HDRx Epic couldn't produce even higher bit depth images. Combining two 16 bit images, even if an "X" image has much less of dynamic range, produces images with much higher dynamic range (18 stops+), which, theoretically, would need more than 16 bit to properly describe it, like open EXR would, I think...
And, yes, 5k size goes a long way toward an ability to produce stunningly detailed images, that you can crop a lot. But I still do not believe, that size of the image has anything to do with dynamic range

 

[Roberto Lequeux]

Well for "actual" DR the talk would be about large photosites vs. small photosites, sensor area, space wasted (if any) between the photosites, electronic noise floor... and about a million other things that probably go into sensor design, so that they don't clip as easily, and so that they are more sensitive, and other what-have-you's that these sensor gurus do. But when you start with a big photo and you down-scale for your final, you do invariably get higher usable dynamic range, because you average out noise and thus lower the noise floor a little bit. So while starting with bigger resolution isn't directly proportional to "actual" DR, you do end up with a little more usable DR than what you have at 1:1 by being able to dig into the shadows a little more. ;-)

 

[Jarred Land]

I validate my point, EPIC is replacing all our Photography Gear!!!

just keep those lenses

 

[Steven Hammer]

Thanks for these amazing images to play with!

 

[Ketch Rossi]

just keep those lenses

He he, count on it Jarred, when I say "I validate my point, EPIC is replacing all our Photography Gear!!!"

I mean all Still Cameras, but we will keep all "L" Series Canon Glass, especially after seen what you guys are up to
with the Electronic CANON Mount, can't wait to shoot it on EPIC

 

[jake blackstone]

Well for "actual" DR the talk would be about large photosites vs. small photosites, sensor area, space wasted (if any) between the photosites, electronic noise floor... and about a million other things that probably go into sensor design, so that they don't clip as easily, and so that they are more sensitive, and other what-have-you's that these sensor gurus do. But when you start with a big photo and you down-scale for your final, you do invariably get higher usable dynamic range, because you average out noise and thus lower the noise floor a little bit. So while starting with bigger resolution isn't directly proportional to "actual" DR, you do end up with a little more usable DR than what you have at 1:1 by being able to dig into the shadows a little more. ;-)

No, I'm sorry, I'm not talking about acquisition. I'm talking about already acquired RAW images processed at different sizes, like you'd originally implied. There is no DR correlation whatsoever....

 

[Roberto Lequeux]

Jake, this is what you quoted earlier, so I am assuming this is what you are referring to:

I tried pushing too hard, but the thing just won't friggin' break at 50%, which is WELL beyond 1080, or 2k! You gotta do something really dumb to get to the breaking point. Starting with 5k REDRAW is a joy.

You actually partially misinterpreting. Though perhaps the last sentence should at least have been on a separate paragraph. I was also still thinking about sharpening, which we talked about earlier. I have been told by many to try to sharpen at the original level, and my (non pro colorist) experiences seem to corroborate that you will get a smoother image, less sharpened-looking.

But that said, you haven't explained why you think the available DR concept is wrong... Perhaps this isn't true with such unbelievably clean images, but say we really underexpose the sensor and start to see some noise... then would you agree with that I described? Some of the noise would get averaged out when you downscale, so while the actual DR would never change, you would be able to dig more stuff out of the darks if need be. This could easily be described as having more usable DR if you are otherwise forced to crush things into black.

 

[Shawn Nelson]

Dang, I can't wait for this camera!

 

[jake blackstone]

Jake, this is what you quoted earlier, so I am assuming this is what you are referring to:



You actually partially misinterpreting. Though perhaps the last sentence should at least have been on a separate paragraph. I was also still thinking about sharpening, which we talked about earlier. I have been told by many to try to sharpen at the original level, and my (non pro colorist) experiences seem to corroborate that you will get a smoother image, less sharpened-looking.

But that said, you haven't explained why you think the available DR concept is wrong... Perhaps this isn't true with such unbelievably clean images, but say we really underexpose the sensor and start to see some noise... then would you agree with that I described? Some of the noise would get averaged out when you downscale, so while the actual DR would never change, you would be able to dig more stuff out of the darks if need be. This could easily be described as having more usable DR if you are otherwise forced to crush things into black.

I'm sorry Roberto, but I don't agree. May be I misunderstood your original point, but grading from 2k version of a 5k image doesn't reduce available DR. There may be theoretical reasons why it may be possible, but in practice, there is no difference whatsoever. Once the image is captured in RAW, it's DR is the same, regardless of working resolution. It's not the same as using new color science, where with use of newer CS DR had been dramatically improved. Crashing blacks to get cleaner image is never a good policy. Proper noise reduction is a much better way to go.
Also, I'm not understanding what "sharpening at the original level" is. For what it's worth, you should sharpen at the very end of the process for best looking images.

 

[Roberto Lequeux]

I usually add at least a touch of sharpening at the very end, and always stay on the conservative side when developing. But don't think that I am trying to tell a colorist what to do. I want you to tell me how you do it, and pick what I like as I learn from you and develop my own style.

About the DR increase, I didn't say you get a huge amount of practical DR. Some have said it is about half a stop or so. It is marginal. And of course I am not suggesting to bury your image to hide noise! I also didn't say that grading from a 2k version reduces available DR... I think I made myself pretty clear. This may be a bit theoretical, and not terribly practical, but it is also true. I am not saying that grading off proper 2k downscaled files is bad. Basically all I am saying that on-line is better. No big secret there.

I am not the artist working hard to grade 90 minutes of gorgeous material time after time. I am a writer director talking about what I do for RAW stills. Is 100% on-line what I would want for my feature? You betcha! But there is no way I am going grade it myself, I know better. And I'll also know better not to feel like I missed out if we can't on-line but still get a great colorist.

 

[Jarred Land]

attached image

something about this one I like... good job Shawn.

 

[Shawn Nelson]

something about this one I like... good job Shawn.

thanks!

wicked good imagery man. I really liked your choices. And playing with it in 5k R3D was fantastic . It feels even cleaner than my MX, which I'm not sure how that would be since it's the exact same chip.

 

[Tim Whitcomb]

thanks!

wicked good imagery man. I really liked your choices. And playing with it in 5k R3D was fantastic . It feels even cleaner than my MX, which I'm not sure how that would be since it's the exact same chip.

60% more resolution, from the same chip, thanks to the new boards, wizardry, et cetera, et cetera.

 

[Bjørn Vermo]

Jake, this is what you quoted earlier, so I am assuming this is what you are referring to:



You actually partially misinterpreting. Though perhaps the last sentence should at least have been on a separate paragraph. I was also still thinking about sharpening, which we talked about earlier. I have been told by many to try to sharpen at the original level, and my (non pro colorist) experiences seem to corroborate that you will get a smoother image, less sharpened-looking.

But that said, you haven't explained why you think the available DR concept is wrong... Perhaps this isn't true with such unbelievably clean images, but say we really underexpose the sensor and start to see some noise... then would you agree with that I described? Some of the noise would get averaged out when you downscale, so while the actual DR would never change, you would be able to dig more stuff out of the darks if need be. This could easily be described as having more usable DR if you are otherwise forced to crush things into black.

Given that your file has a greater pixel depth than what the camera is delivering, then you will theoretically be able to gain one more bit if you reduce your resolution to half the number of pixels. In real life, obviously, the gain is smaller.
If you have clipping, obviously nothing will be able to help you, but if the noise floor is what limits your range then you can trade resolution for depth.

 

[Vishal Prabhakar]

My grade

My grade

My Grade 1

Thanks Again )

 

[jake blackstone]

Given that your file has a greater pixel depth than what the camera is delivering, then you will theoretically be able to gain one more bit if you reduce your resolution to half the number of pixels. In real life, obviously, the gain is smaller.

If only this statement had any base in fact. Unfortunately, if you express an 8 bit material in, say, an Open EXR, you will still have an 8 bit material. All you just end up with is 24 empty bits. Sorry, but this approach doesn't work...
And for the record, the file in question is 16 bit deep and so is the material from the camera.

 

[Bjørn Vermo]

If only this statement had any base in fact. Unfortunately, if you express an 8 bit material in, say, an Open EXR, you will still have an 8 bit material. All you just end up with is 24 empty bits. Sorry, but this approach doesn't work...
And for the record, the file in question is 16 bit deep and so is the material from the camera.

It has worked for me, no big difference but sometimes enough to make it worthwhile.

You would be right if there was no such thing as noise. To make an easy example, imagine a monochrome 8-bit camera recording values from 0 to 255. Now, let us get that into a 10-bit file. We end up with a range from 0 to 1023, but every pixel will have the two lowest bits set to zero. Nothing is gained yet.

The dark areas from this camera will have some noise. Some of this noise is strong enough to make a pixel digitize as the next higher value. It is more probable that this happens if the light for that pixel is almost up to the next level than if it is almost down to the previous level.

Let us assume that a uniform area of the image would have yielded a quad of pixels with the value 8 in the absence of noise, and that the same area would have been recorded as 9 with a better (native 10-bit) camera. Because of noise, we get the pixels 8,8,8,12.
Then we scale the image to half the linear size, so our four pixels get binned together as 8+8+8+12 = 36 divided down into 10-bit space giving us 9.
If the area was slightly brighter so that the better camera would have recorded 11, statistics would make it more probable that our 8-bit camera with noise would yield a quad of 8,12,12,12 which would give us a pixel value of 11 after reduction.

Practical scaling algorithms are more complicated, but it still works. I have improved the gradation of old (early CCD) tapes by importing into 16-bit space and then downressing. Sometimes, slightly better gradation and less noise is more important.

 

[jake blackstone]

It has worked for me, no big difference but sometimes enough to make it worthwhile.

You would be right if there was no such thing as noise. To make an easy example, imagine a monochrome 8-bit camera recording values from 0 to 255. Now, let us get that into a 10-bit file. We end up with a range from 0 to 1023, but every pixel will have the two lowest bits set to zero. Nothing is gained yet.

The dark areas from this camera will have some noise. Some of this noise is strong enough to make a pixel digitize as the next higher value. It is more probable that this happens if the light for that pixel is almost up to the next level than if it is almost down to the previous level.

Let us assume that a uniform area of the image would have yielded a quad of pixels with the value 8 in the absence of noise, and that the same area would have been recorded as 9 with a better (native 10-bit) camera. Because of noise, we get the pixels 8,8,8,12.
Then we scale the image to half the linear size, so our four pixels get binned together as 8+8+8+12 = 36 divided down into 10-bit space giving us 9.
If the area was slightly brighter so that the better camera would have recorded 11, statistics would make it more probable that our 8-bit camera with noise would yield a quad of 8,12,12,12 which would give us a pixel value of 11 after reduction.

Practical scaling algorithms are more complicated, but it still works. I have improved the gradation of old (early CCD) tapes by importing into 16-bit space and then downressing. Sometimes, slightly better gradation and less noise is more important.

Great.
Just show me a 4k image in comparison to an extracted 2k, that has increased DR by a 1/2 a stop or exported RAW as an Open EXR, that will result in extra 1/2 DR and then will talk

 

[Vishal Prabhakar]

My grade 2

My grade 2

My Grade 2 )