EPIC MX Highlight Rolloff thread....

[Luis Otero]

Thanks Luis,

If all of this is true and the author has their original R3d, the good news is that the footage can be re-graded, and an easier transition is likely possible. This footage can yet be improved by a knowledgable Colorist if the desire exists.

That is why I have ask people to post a link to the R3D. So far, only person one has done it.

 

[Bob Gundu]

ProRes 4:2:2 is 10bit
ProRes 4:4:4 can be up to 12bit

Bold statement, but you got it 100% right, my friend. That is one of the most important differences that separate decent from excellent DPs.

Tell me about it... I see any kind of footage brought to me for grading, and everyone, even if the footage is 8 bit , wants it to look like film. I mean, there is so much that can be done to emulate a style, but we are not magicians.

BTW, ProRes HQ footage is 8 bit. Even some .mov uncompressed files are 8 bit, depending on what software was used to generate them (FCP 7 and Compressor generates 8 bit, but Resolve generates it at 10 bit). However, ProRes 4444 is 10 bit.

 

[Gunleik Groven]

That is why I have ask people to post a link to the R3D. So far, only person one has done it.

Understandable.

I donot mind sharing testfooty (and have done that a lot over the years.
Stuff which is work, is not relevant to share, though, and I do little pure testing these days...
Except from when prep'ing productions for others or when I teach, and in those cases, the footy belongs to my employer...

 

[Luis Otero]

ProRes 4:2:2 is 10bit
ProRes 4:4:4 can be up to 12bit

Bob,

I disagree with you, and this is the fact, and why FCP 7 and Compressor generate 8 bit ProRes HQ. Resolve and other software is a different story. If you check the parameters before encoding, it says 32 bit, and we say GREAT, 12 bit per color! It does not mean 12 bit per color channel; it means 8 bit per color channel and 8 bit for the alpha channel.

I used to think the same way, but after a lot of research I found the explanation of why I was getting banding from ProResHQ processed from FCP and Compressor.

You will get 10 bit Uncompressed .mov from Resolve,, but not from FCP 7 or Compressor. Check the parameters of the file and you will see that big 32, but it includes the 8 bit alpha.

Edit note: ProRes always will include the 8 bit alpha channel. So, if you deduct 8 from the 32, it will give you 24 bits, which are equally distributed to each RGB channel. For PR4444 is the same logic. An 8 bit alpha channel. Resolve ask if you want RGBA 10 bit or RGB 10 bit. So the difference is with or without alpha channel, hence it truly renders a 10 bit .mov.

 

[Luis Otero]

Understandable.

I donot mind sharing testfooty (and have done that a lot over the years.
Stuff which is work, is not relevant to share, though, and I do little pure testing these days...
Except from when prep'ing productions for others or when I teach, and in those cases, the footy belongs to my employer...

That is understandable, Gunleik. Same here.

 

[António Fagundes]

So clipping is due to uncoated lenses and sRGB?

 

[Daniel Stilling]

Possibly. But mostly due to wrong choices of tools or wrong usage of those tools. pretty simple actually...
This is such a subjective issue that can be made as complicated or as simple as you like. If you prefer how the Alexa does it, that should be your tool of choice. If the Epic is not doing it for you, it's the wrong tool.

 

[Luis Otero]

Hey everyone,

Just found a pair of examples of before and after Color Space transformation. Graded for DCP, and transformed to sRGB for web posting. Different angles, but same set-up. Shot on Red One.

This shows how the sensor handled the overload/clipping, but how the transformation created the artifact.

 

[Gunleik Groven]

But in there lays what to me is one of the biggest differences between R1 MX and Epic...

(it is fixable on the R1, but I have a hard time trigging it in a bad way on the Epic)

 

[Luis Otero]

But in there lays what to me is one of the biggest differences between R1 MX and Epic...

The point that I am trying to make with both images is that even with the Red One sensor, without Color Space transformation, the sensor registers as Pure White the clipping/overload. And the culprit for the magenta artifact is the transformation.

 

[Scott Crawley]

Redcode does not presumably use this aspect of the signal to compress the data.

Sorry for quoting myself but I have been thinking about this and want to revisit it. In my ignorance I am having a hard time thinking of a way my previous remark could be accurate.

I know nothing about Redcode or other file compression schemes except that they work by discarding data. It seems logical that the bigger the pile of bits, the easier it is to remove some without anyone noticing. How do you do that successfully? Two thoughts come to mind.

1. If you think of the image in terms of gradients and points on a curve as I have been, and consider removing only points on a line or a continuous gradient then those points will be sorely missed. If you think of the image in terms of a 3D LUT, as the voluminous mass that they are then it becomes easier to sneak some pieces out without anyone noticing.

2. When i think about removing data like someone trying to hide in plain sight I think of camouflage. Camouflage is successful because it breaks up the visual lines of an otherwise recognizable form. In other words it reduces it's linearness. Therefore, the less uniformly you remove the data... The less linear your method... The less it will be noticed.

I suppose compressionists like Graeme think about this shit all day. I suppose that is why some mathematicians love chaos. Graeme, do you love chaos? ;-)

 

[Gunleik Groven]

The point that I am trying to make with both images is that even with the Red One sensor, without Color Space transformation, the sensor registers as Pure White the clipping/overload. And the culprit for the magenta artifact is the transformation.

understood, just saying that the two are not identical.
And that is part of the discussion, too. Which cam behaves how.

When the Epic came, we were wondering if we

1) should stick to MX
2) shoot mixed MX/ Epic
3) shoot epic

for the next larger series (6 months of shooting)

The difference in behaviour under overexposure was one of the main reasons to go Epic. Also how they behave at underexposure is quite different.

Point is:
Yup, they both overload, but the resulting image at over- and underexposure differs quite a lot.

Epic it was...

 

[David Battistella]

Great post here:

http://www.reduser.net/forum/showth...!!!-(PART-2)&p=1242299&viewfull=1#post1242299

We learn something new every day on REDuser...


David

 

[António Fagundes]

Just a little fun here...

So in the thread David mentioned above, there's a link to a high contrast and color temperature test between several cameras, including the Red Epic, Arri Alexa, BBCC, Canon C500 and Sony F55.

I decided to download the DPX and the R3Ds provided on the site and check out their luma values and curves. I did this in Photoshop CS6 and exported a tiff from Redcine X to Photoshop in order to have all the camera's files in one place.

I didn't want to do anything fancy, just check the curves and see how those highlights were behaving in all of the cameras (comming to think of it I forgot about the F55...)

To be completely honest here I think the Alexa had the better highlights and color rendition. As soon as I saw the Red file open in the Redcine X I saw the highlights were clipping straight away. There's a yellow color cast all around the highlights from the door to the windows.

After I checked the curves (which by themselves are very revealing as well) I did a very basic black and white points adjustment just so to see how each camera got their colors. I think the Red, due to its yellowish color cast, transfered to a very greenish/yellowish color, whilst Alexa and BBCC performed quite well to be honest. Clean colors. To my eye at least.

Note that I didn't do anything different in any of the cameras, did exactly the same thing in all of them, and of course they all gave me different results... but look for yourself down at the pictures. Oh, they were all 16bit tiffs and my Photoshop settings were at RGB, 16bit.

Actually the Canon C500 performed horribly!

First the untouched frames and their histograms...

RED EPIC

Notice how steep both ends of the histogram are, especially the right side (the highlights) there's no smooth transitions there.... this is RedGamma 3 with RedFilm on Redcine X.


ALEXA

Highlights don't seem to be clipped and the histogram seems to be 'fatter' compared to the Red which is all falls down bellow the mid point of the scale. Alexa's histogram is not only tall but wider as well.


CANON C500

Very contrasty image right off the camera. Highlights seem to be clipped as well. Although the histogram looks quite good, the image itself is not that pleasant, not to my eyes at least and not to an image that is supposed to be 'flat' and good for grade. This looks too contrasty already...



BBCC

Good enough image in my eyes, but the histogram is also kind of low, notice also the clipping highligh on the histogram, a straight line all the way up. Actually, the Alexa seems to be the only histogram that shows no clipping.

It's not me saying, it's what the histogram shows...

 

[António Fagundes]

Some very basic black and white sampling to give the image contrast. By doing this we bring the natural color cast the image. I think the Red file is the most pronounced. Due to the yellowinsh clipping.


RED EPIC

ARRI ALEXA

CANON C500

BBCC

 

[Björn Benckert]

Some very basic black and white sampling to give the image contrast. By doing this we bring the natural color cast the image. I think the Red file is the most pronounced. Due to the yellowinsh clipping.


RED EPIC

ARRI ALEXA

CANON C500

BBCC

To me you need a really untalented DP to shoot such shitty shot with an epic... and you need to make sure not to do anything to balance the shot before you grade it. why keeping the highlight highly green / magenta tinted when you got a raw image? And the epic can shoot that shot with perfect DR for the interior and exterior and then get a grade that blows the other far out of the water using the HDRx.

To me it's kind of like you would throw a film camera in there but instead of using the full potential of the film you took a grab from the video tap and then told us it does not have any dynamic range. Use each camera to the full potential or the test is not really valid.

I saw another test where they compared sharpness and such, then they shot the epic in 4k and the comparison was done in done in HD... this test kind of does the same.

 

[luigivaltulini]

Antonio...
the guy has completely the wrong exposure with Epic ...
For example:
If you control the movies at night the F55 and calibrated to 1250 and Epic has a base ISO of 2500!!!
What the fuck has closed the iris!!
the F55 you have to expose it to almost 5000 iso how to get to Alexa 800 iso!!
And the adding alexa better than an F55 in terms of noise!​​!
And EPIC turns out to be half the signal .. necessarily be better!

Not go ahead otherwise take anger ...
And you trust someone like that?

do not make comparisons with these videos please ...

 

[David Battistella]

I looked at these. I followed the exact same method.

When I processed the R3D at 320 in RCX and did a white balance on the the image it had a wider DR. I was surprised that the WS frame was so out of focus. I did not see a lighting diagram so I don't know if that was sunlight or a source outside the window. The lighting or T Stop is clearly not consistent in the two shots. In the MS the window holds th highlights with no problem. WS, not so much, but that is what I would expect from a test like this.

But I see how this test would help illustrate what you are trying to say.

Both DPX from R3D with WB and exposure done prior to export as RED recommends.

David

 

[Robert Ruffo New]

But in there lays what to me is one of the biggest differences between R1 MX and Epic...

(it is fixable on the R1, but I have a hard time trigging it in a bad way on the Epic)

True that!

 

[luigivaltulini]

I found some time ago strokes taken directly against the sun to test .. what do you think?
used only REDCINE-x

[/URL][/IMG]

[/URL][/IMG]