Resolution concerns, 2K 4K?

[Álex Montoya]

Yeah, the blond girl's one is the most beatiful pic I've seen from the RED.

You may see the resolution difference better here:

Though this picture kinda preoccupies me. It's not the blown-out window but the harshness of the transition that I don't like.

 

[Justin O'Neill]

I think the issues you are seeing with the footage from the Blair, Tonaci, Evin, Brook crew, are simply due to the footage not being 100% in focus.

This is entirely understandable when you consider their situation. Blair's RED LCD and EVF were not ready yet so they were using Evin's Panasonic monitor for focusing. Brook mentioned they were zooming in at not quite 1:1 and judging focus with their eyes on a 8" screen. Also Gaeme's magic pixies aren't enabled yet.

They were out in the field sharing their learning experience with us in realtime. I would be surprised if the images coming back were 100% technically correct.

 

[Jack Wester]

Graeme,

I agree, but resampling 4x4 pixels to a single pixel would effectivly and beautifully get rid of any aliasing as well as reduce some noise, right? So when can we have a 16K red camera? Just kidding. IMHO, I don't think we need more resolution at all. My problem would be being out of focus all the time. I would never reach the limits of this camera without some serious help with fine tuning focus.

In your opinion, given a really good dialing in on the focus, how important is the lense to get really sharp details for the resolution of your sensor?

 

[Michel Hafner]

And as I pointed out, there are vast differences in how the RAWs are developed.... And that we use a OLPF designed to stop aliassing, which is totally unnaceptable in a motion picture,
Graeme

I agree, but digital intermediates of past and present often have aliasing nonetheless. :angry03:

 

[Graeme Nattress]

No, resampling does not get rid of, and cannot get rid of aliassing. Once aliassing is in the image, it's practically impossible to get rid of and downsampling does not get rid of it. Correct downsampling can only stop aliassing occuring - it cannot remove it.

Getting the lens sharp is not perhaps as important as getting the MTF high in the lower frequencies which govern overall image contrast. This means good glass with good filtering. Maximum resolution, for me, is not the overall lens requirement - "look" is.

Graeme

 

[Graeme Nattress]

I'm a founder member of:

1) the "Aliassing is bad" club

2) the "linear means a straight line" club

Graeme

 

[Deanan]

There are a few unknowns for the first images that make it hard to judge
the image for sharpness like...
What is the sweet spot of the lens vs what aperture it was shot at.
How steady was the camera or subject?

It's also important to note that the RED one was designed to produce beautiful images that are practical to acquire rather than hyper sharp images with great looking specs.

A 4k camera designed to produce hyper sharp images will have endless problems with aliasing and will make it a giant headache to pull focus because the perceived depth of field falloff is much steeper.

Deanan

 

[Graeme Nattress]

Not least to add that edge sharpness contributes towards the perception of motion judder - something you don't want to amplify when shooting at 24p or 25p.

Graeme

 

[Karl H]

Darkline, if only it were so simple.....

The measured resolution from the 4k is over 3k, but say we downsampled it to 3k, we'd now need to be in RGB, not raw, and therefore the file size just got bigger, not smaller. Downsample to 2k, and you've got a 2k full of nearly 2k resolution, but the file size is 3/4 of that of the 4k and you've lost 1k of resolution for that. That's why 4k is best recorded as 4k.

Graeme

gotcha.... but didnt Jim mention that you had found a way to do scaled 2K raw? So maybe scaled 3K raw too?

 

[Deanan]

I'm with Graeme.

Aliasing is true evil. Even if you took a 4k image with aliasing
and downsampled it to 2k, the aliasing would still be just as offensive.

DI's I've seen that have aliasing are generally because of shortcuts in
the workflow and excessive sharpening prior to filmout.

Deanan

 

[Lauri Kettunen]

If you put a > 4k resolution into a 4k sensor, you get aliassing, and that's nasty and not removable by downsampling. That's why we only get >3k out of the 4k as we're low pass filtering the image.

Graeme, would you agree with me that in the end of the day this, let's say, "lost of data" becomes a competetive edge of the RED sensor? I mean, since you anyhow have to add the low pass filter, that's also a reason why the r,g and b sensor pixels need not to be above each other, but instead it's enough that they are side by side.

This also explained why you can save so much of the original information in the deBayer process. It's a sort of optimal compromize making a very good use of the fact that the aliassing filter has to be there. In terms of the discussion we had in the other thread, the filter guarantees the data is "somewhat regular" or should I say "smooth" as it smoothens just a tiny but necessary amount of any kind of "fractal details".

Quite many on this forum also have wondered about the lacking noise. Graeme, to which extent the wavelet compression has an effect to this? Is it another place where the necessity is found as an advantage by the users? In my eyes the aliassing filter, the sensor, and the compression all work together in a saddle balance such that the necessities do not end up as being compromizes of quality. Finaly, it's almost like a jackpot that one wins also with the lenses; We all know how expensive it is to design high quality lenses to HD 3 CCD systems.

Still don't understand, how come the big manufacturers missed all this and "got surprised the trouses on their ankles" as we use to say in Finnish?

 

[Graeme Nattress]

Actually, to my eyes, aliassing gets more offensive on a downsample!

Graeme

 

[Graeme Nattress]

Engineering is always a series of compromises. I think we've taken those that lead to a very good, high resolution, low file size image.

Graeme

 

[Evan Owen]

I'm with Graeme.

Aliasing is true evil. Even if you took a 4k image with aliasing
and downsampled it to 2k, the aliasing would still be just as offensive.

DI's I've seen that have aliasing are generally because of shortcuts in
the workflow and excessive sharpening prior to filmout.

Deanan

Deanan (or Graeme), most RED footage that I've seen so far has been great in this regard–just very, very smooth. But going back to what Concrete mentioned, I too have noticed that areas with blown-out highlights often have very hard edges, even to the point of being badly aliased.

Is this something that will be worked out in the future?

 

[Graeme Nattress]

I think the output sharpening was turned on for that shot, and perhaps it should not have been.... Without access to the raw file, I cannot guess further. I'm working on "better" "nicer" output sharpening for those who need it though.

Graeme

 

[Jack Wester]

No, resampling does not get rid of, and cannot get rid of aliassing.

I feel the knowledge and experience I had about aliasing is flying out the windows And I'm a 100% sure that you're right, as it would be presumptuous and arrogant for me to believe otherwise. So this simply means that I've got things to learn.

From my days as a graphics programmer drawing vector graphics, the predominant aliasing effect was getting jaggy edges and anti aliasing ment getting rid of those jaggy edges. That kind of aliasing would definitelly go away when outputing to a higher resolution and then downsizing using averages of 4x4 pixels to a resulting in a single pixel. Expensive but effective. But as aliasing is a concept valid in signal processing in general, my view on aliasing does not seem to be valid to the world of sensors and sampling, so I'll just shut up and listen on that matter.
oster_stupid:

EDIT - I Found this definition on Wikipedia for the kind of simplistic antialiasing I often used in my programming. Not to prove you wrong, because I honestly don't think I really understand the aliasing problem you're talking about and that I'm at fault here

"...anti-aliasing by supersampling usually means that each full frame is rendered at double (2x) or quadruple (4x) the display resolution, and then down-sampled to match the display resolution..."

 

[Evan Owen]

I think the output sharpening was turned on for that shot, and perhaps it should not have been.... Without access to the raw file, I cannot guess further. I'm working on "better" "nicer" output sharpening for those who need it though.

Graeme

That might be the issue there, but the rest of the image doesn't look sharpened at all, to my eye anyway. It looks silky smooth – except for the edges of the blown-out highlights.

Thanks for your work on the sharpening algorithm though, I'm one who likes careful (or no) sharpening. :biggrin:

 

[Evin Grant]

The pudding...

This 100% 4K grab was sharpened using smart sharpened in photoshop, then the image below was down sacaled to 2K and then upscaled back to 4K.
I think the result speaks for it's self. Look especially at the fine detail in the shirt stiching. That being said movement is a big factor, anything moving will not achieve this level of detail, but it really does'nt have to.

Tiff
http://www.reduser.net/evin/4k2k.tif

 

[Karl H]

Yup I can see that clearly.

what ISO and shutter speed was this taken with Evin? there is much less noise than in the indoor shot

 

[Graeme Nattress]

Yup, that's all you can do for drawing digital imagery - plot at a much higher resolution and average, or better still calculate the percentage of the line passing through a pixel and plot it that brightness.

However because aliasses are frequency mirrored artifacts, they fold back into the data in such a way that they can't be removed by downsampling techniques. For downsampling, basically you low pass filter the data, and then discard any data not needed to create the new downsampled image. So for 4k to 2k, you'd filter with a filter that removes any frequency above that which 2k can sustain, then discard every other sample. Because aliasses fold back into the data, the contaminate the full frequency spectrum, so no filter bases approach can remove them without removing real information also.

Graeme