David Stump's RED files are up

Status
Not open for further replies.
they are getting uploaded slowly one by one.. David has given permission to quote his notes verbatim, so this should be fantastic.
 
Yay! Downloading the first one now ... 1.3MB/sec .. school has a nice connection. :) EDIT (20 sec later): Done!

Did I understand that as permission to host mirrors Jarred?

- Mikko
 
Well I figured I'd try to colour correct the F11 one to see how close I could eyeball it (on a Dell 2407WFP so don't be TOO critical!):

RGB_f11_cc_sRGB_8bit_small.jpg


I added in my own RGB chart bottom right corner (sRGB profile) to match with and I'm pretty happy with the results for a quickie.

I did notice a few artefacts on the original image which no doubt everyone will spot as well:

- dust on the sensor or lens in a few spots
- on the CamAlign chart the right hand line resolver seems to have one vertical line where it "jumps" a line of pixels
- on the in camera GretagMacbeth chart the right hand black level appears to have some sort of yellowish horizontal distortion on it?

Anyway, great to see a professionally shot image (I know it's a test with stops etc) and great to see what can be done with it!

If anyone wants a full size version of my CC'd image you can download it here (1.7MB 8bit JPG).
 
I love how even without the chart filling the screen, you can see every single line to the ends of the resolution trumpets! :shocked:

They need to invent a sharper printer for charts next!

- Mikko
 
interesting behavior at high spatial frequencies

interesting behavior at high spatial frequencies

My starting point was the f11 TIFF image at
http://www.cinematography.net/Red/ColorChartsExposureSeries/RGB_f11.tif

The image looks so nice with no visible flaws, I had to push really hard to find anything unusual. After sharpening to a completely ridiculous degree in Photoshop and blowing up to 200% using nearest-neighbors, you can see an interesting effect in the PNG crop below. There is relatively more single-pixel-level noise where the original image has high spatial frequencies, at the center of the starburst. I don't know if it is the debayer algorithm or a denoise or hot-pixel-removal process. I can't think what else would give you that kind of selective response (very smooth everywhere except in the sharpest-detail regions). Note, this effect is completely invisible to me at 100% with no sharpening.

Red-f11-RGB-sharpened-2x-detail-star.png


I should add that I have seen very similar things to this on DSLR images of resolution charts, near the point where the lines converge to the resolution limit. You can go to dpreview.com and see examples, for example from the EOS 1Ds Mark II (16 Mpixel)
http://www.imaging-resource.com/PRODS/EOS1DS2/1DS2DBWBD.HTM ...and of course if you did a 4k scan of any normal cine film stock and did similar sharpening and enlargement, it would be pretty noisy everywhere.
 
Last edited:
Even when the chart fills the screen, the camera out-resolves the chart. I think we need a new chart :-)

The pixel alignment and funnies in the black are a problem with Frankie, and why we don't move her. David says ". I must also state that in the course of dollying "Frankie" from one setup to another I ran over a cable which caused a sync issue in the signal for part of the exercise. We restarted the camera and the issue went away, but I am using some frames from the series of grayscale and color chart brackets we did before the reboot."

Obviously, those kind of problems won't exist in the RED camera, and remember Frankie is a sensor test rig.
 
Funky single pixels like that come from a number of things:

1) uncharacterized sensor

2) blowing up nearest neighbour and over-sharpening

3) above interacting with adaptive demosaic.

Yes, such things are normally invisible until you go silly with them! Try the same thing with any camera and you'll get serious nasties.

Graeme
 
Yes.. thanks for making that clear Graeme.

When your looking at the frame, the far right slice, about 20% of the frame, is out of alignment and has a broken signal. At the time, it was decided that since it was a latitude test and that the slice did not affect the charts, that we would still be able to use the shots to judge latitude so it was ok to proceed instead of just calling it a day. The Greenscreen shots should be clean of that defect.
 
Presumably when the line frequency of a pure black and white image approaches the pixel frequency of the Bayer pattern, no de-mosaic algorithm in the world is going to produce a perfect b/w image, and some off-colour pixels are inevitable. Over sharpening will obviously exagerate this.

With real-world images, in a real-world viewing situation I can't see this being a problem at all.

Nick
 
Yes, it's a pathalogical image to an extent, but even still, under every normal viewing condition I've not had any bother with rez charts yet. When I did over-sharpen and stuff on the image above, I couldn't make it break that way :-)

Graeme
 
So has it been determined how much latitute Frankie has? I realize all the images and notes aren't up yet so maybe those in the know are waiting for David Stump to reveal his results without giving anything away first.
 
Is there a well-defined way to determine an exact latitude? Seems like it must include some subjective judgement about what image quality is "good enough". Particularly in the blacks where in theory you can push them up almost any amount, but at some point the resulting image is going to be too noisy. So you might end up with many different effective latitude numbers, depending on if your final output is 4k, 2k, HD or SDTV and they might be considerably different.
 
I must admitI have never had that much fun examining something as dull as a chroma-chart before.

Gunleik
 
Status
Not open for further replies.
Back
Top