Welcome to our community

Be a part of something great, join today!

Register 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 :)

Red Resolution test

D

Dunkan Wolf

Well-known member
Joined
Dec 2, 2013
Messages
330
Reaction score
0
Points
0
interesting stuff

https://tech.ebu.ch/docs/tech/tech3335_s16.pdf

4k mode (Fig.1, 3840x2160)There is little image-content resolution above about 2450x1380, with unacceptable coloured aliasing athigher frequencies. Aliasing is also present in the smaller double-frequency pattern, centred at 7680(horizontal) and 4320 (vertical). This shows that the optical low-pass filter is not eliminating these unwantedhigher frequencies. This performance is very like that from a single 1920x1080 sensor for conventional HD,for all the same reasons. The ‘Dragon’ decoding is slightly better in that it appears to deliver a little moreresolution, but the level of aliasing is also higher, so there is no net gain. This mode is not acceptable for 4kshooting.

5K mode (Fig.2, 4800x2700)This is a little better, but not much. The image-content resolution limits are about 2930x1650 with the samelevel of coloured aliases above those limits, but centred on higher frequencies. Again, the Dragon decoderdelivers a little more resolution and aliasing.

6K mode (Fig.3, 5568x3132)This is a little better still. The image-content resolution limits are now about 3400x1900. The level ofaliasing appears to be lower, because the optical low-pass filter and lens are not passing so much of theunwanted frequencies to the sensor. This mode is acceptable for 4K shooting, in the same way that shootingon a 2880x1620 sensor is good enough for conventional HDTV. However, the zoom lens did not fully coverthe sensor, resulting in significant vignetting at wide angles, clearly the sensor is rather larger than ideal forthe design of the lenses.Although not specifically tested, it is safe to say that the 6k mode is also the best way to get HD footage,since aliasing is at its lowest. Certainly, the 2k mode will not be good enough
Click to expand...
 
From reading his report on the ARRI Alexa for Tier 1 HD broadcasting also, I'd say he has pretty high standards so I wouldn't take this Red Dragon report as being overly critical, you have to figure that only the Sony F65 is in the same ballpark as the Dragon for delivering a 4K image that meets his standards (until he tests the Alexa 65mm...)

https://tech.ebu.ch/docs/tech/tech3335_s11.pdf
 
@ 6k (5568x3132) they get 3400x1900 resolution limit. This is only 61% of the resolution due to the bayer process. Some time ago we where talking of 70% on the MX.

6K mode (Fig.3, 5568x3132)
This is a little better still.
The
image
-
content
resolution limits are now about 3400x1900. The level of
aliasing appears to be lower, because the optical low
-
pass filter and lens are not passing so much of the
unwanted frequencies to the sensor. This mode is acceptable for 4K shooting, in the same
way that shooting
on a 2880x1620 sensor is good enough for conventional HDTV. However, the zoom lens did not fully cover
the sensor, resulting in significant vignetting at wide angles, clearly the sensor is rather larger than ideal for
the design of the le
nses.
Although not specifically tested, it is safe to say that the 6k mode is also the best way to get HD footage,
since aliasing is at its lowest. Certainly, the 2k mode will not be good enough.

 
It is interesting that he used a zoom lens that did not cover the whole sensor. The limitation could have been the lens itself. Also - unlike with the ARRI test - he did not specify which lens he used. Why not use a known high-resolution prime which would eliminate this possibility?
Other then that - very interesting reading indeed...

:sifone: Peter

PS: David beat me to it... ;o)
 
compression is a big factor too... the more compression, the more likely its gonna be softer. That being said I have delivered a decent amount of content at 4k shot in 3k and 4k on my scarlet in their lowest compression ratios and it looks pretty dang good on a 60 inch 4k display.
 
Interesting read and I'm finding a few variables that need to be locked down a bit as my results using similar methods vary in accuracy compared to this by a smidge.

I need to know precise lens information here. T4 isn't what should be used with the lens described for this particular test in regards to resolution. Additionally he is testing HD formats on Dragon which is what the spec is looking for, but I believe he should also test shooting 6K FF with a 1.78:1 full height extraction as there's more resolution to be found. There's also the missing concept of different scaling algorithms being utilized. I have some questions regarding the DR measurement as well, but he's relatively close to "seeing" everything, just not clear on the measured results or test method.

Locking down and testing a few of those concepts you'll find that Dragon is more Tier 1 than this test reveals.

I'm working on something I can show post-NAB regarding tests like these for all cameras that should make "life simple" for testing and evaluating.
 
Is there any resource for the different scaling/debayering algorithms? i.e. what they are doing mathematically and what kind of artifacts they might cause. There's a good wiki page about them but it doesn't seem to cover them all, not even the ones to be found in RCX.
 
Alan is a very competent person, I had to do it several times. Many cameras Sony, Thompson, Ikegami. for broadcast have its setting, are wonderful.
I did several courses at Pinewood in London with him, wonderful crazy. I love it.
I know I had to do these tests with Dragon and Alexa at year end 2014, Li was expecting.
Now go read on ftp all the tests, I'm curious to see beyond the resolution, difference dynamic range between Alexa and DRagon thing to have measured.
 
Last edited:
Beside the obvious issue of the lens choice, one thing confused me a bit. He states "a standard HDTV test card was framed to fit exactly half the horizontal and vertical image dimensions, thus simulating the condition in which camera footage is scaled to 3840x2160 after shooting." I am not sure why this was done since, it seems to me, the best way to get to HD is by shooting 4,5, or 6K and downsampling in post. Why not frame the chart at the full window of the recording?
 
Patrick Tresch said:
@ 6k (5568x3132) they get 3400x1900 resolution limit. This is only 61% of the resolution due to the bayer process. Some time ago we where talking of 70% on the MX.
Click to expand...
Like most tests there is always room for subjectivity. The only way to get to true resolution is with uber oversampling. That's true of any and all sampling whether it's a microphone or it's camera sensor or lidar or a 3D Render or a film scan. Most 3D Renders start at 400% oversampling and go up from there (It's one of the reasons why actual 3D rendering time isn't *that* much of an increase for 4k). Take off the OLPF and I bet you'll get a lot closer to the theoretical efficiency of debayer but then you get artifacts. Even at 6k for 1080p I still sometimes see artifacts. Where you see sufficiently few artifacts to satisfy your own personal standards of "Artifact free resolution" is pretty subjective. As David said this test is pretty demanding so you shouldn't see this as Dragon having less resolution than MX, just that his measurement techniques have a higher cutoff for 'true resolution'.

David Mullen ASC said:
He might have squeezed a bit more resolution out of a Master Prime instead of a zoom, who knows...
Click to expand...

It sounds like his largest limiting factor wasn't actually detail but aliasing at said detail levels. If anything he's critical of RED for aliasing not for being insufficiently sharp. My guess is that a Master Prime wouldn't improve his results if anything they would reduce his rating for revealing even more aliasing.

Terry VerHaar said:
Beside the obvious issue of the lens choice, one thing confused me a bit. He states "a standard HDTV test card was framed to fit exactly half the horizontal and vertical image dimensions, thus simulating the condition in which camera footage is scaled to 3840x2160 after shooting." I am not sure why this was done since, it seems to me, the best way to get to HD is by shooting 4,5, or 6K and downsampling in post. Why not frame the chart at the full window of the recording?
Click to expand...


Because it was a 1080p chart. So if you frame it up to half width it becomse a "2160p" chart since it's only filling half the frame. The resolution lines are twice as dense in frame and therefore tests the resolution better.

As to the vignetting and lens not filling the frame... no problems there since he's only using the middle half. That's good testing procedure since you don't want to measure resolution at the edges of frame anyway. Good technique isn't to count lines across the entire frame, it's to count int he center of frame where the image is sharpest and then extrapolate.
 
Just chatted a bit with Alan, apparently they ditched the zoom described here and utilized approximately a 70mm prime. i'll post up some results after NAB.

Nice guy. Shared a couple notes on where I "went different" against what's presented here as there are ways to achieve "more".
 
Gavin Greenwalt said:
Because it was a 1080p chart. So if you frame it up to half width it becomse a "2160p" chart since it's only filling half the frame. The resolution lines are twice as dense in frame and therefore tests the resolution better.
Click to expand...

Thanks, Gavin. I think I get the part about a better test procedure. Looking past that to "real world" use, might it be true that framing the same chart at full frame (shooting at 5K, for example), then downsampling to HD, would actually "resolve" better for the HD output?
 
Phil Holland said:
Just chatted a bit with Alan, apparently they ditched the zoom described here and utilized approximately a 70mm prime. i'll post up some results after NAB.

Nice guy. Shared a couple notes on where I "went different" against what's presented here as there are ways to achieve "more".
Click to expand...

Phil, i think is same test of Alexa:

https://tech.ebu.ch/docs/tech/tech3335_s11.pdf

So also the dynamics range near Dragon.

Thus the dynamic range explored for this test must be about 11¾ stops, so the claim of 14 stops seems entirely reasonable
 
Terry VerHaar said:
Looking past that to "real world" use, might it be true that framing the same chart at full frame (shooting at 5K, for example), then downsampling to HD, would actually "resolve" better for the HD output?
Click to expand...
If it's a 1080p chart all you would be able to learn from the 6k Dragon sensor would be that it "Resolves at least 1080p" which one would hope is true/obvious. :D



Phil Holland said:
Shared a couple notes on where I "went different" against what's presented here as there are ways to achieve "more".
Click to expand...

Make sure he blackshades. The fact that he changed the shutter duration on the dynamic range test made me a bit ancy. Ideally you would only use ND or Aperture for a dynamic range test and leave the sensor static since it's the device being tested. Also something that Stacey and I learned when doing our DR test was to be sure to use the adaptive fan because the noise characteristics can change a lot from second to second if the temperature varies at all. Here's to hoping that Weapon has a peltier or something else that can better regulate temperature than just copper.
 
For anyone interested - here is the same analysis for MX Epic...:

https://tech.ebu.ch/docs/tech/tech3335_s04.pdf


Some "highlights":

"At gain settings higher than about +16dB (about 5000 ISO) the noise level is quite visible, and could be disturbing, being typical of small, consumer HD camcorders with ⅓” sensors. At -10dB (250 ISO) the noise level is impressively low."

"And it means that, when shooting for 1920x1080 TV output, the aliasing can be reduced dramatically by using slightly softer lenses and setting the camera to 5k or 4k mode, since the down-conversion in RedCine uses the Lanczos algorithm, which appears to be the best option available at present."

"The RED Epic is different from other cameras. It deals with images differently, and has to be treated differently to get the best from it. When set to 5k HD or 4k mode, it makes very nice pictures with content up to about 3.7k, but when making 2k pictures directly it is a little disappointing except that the frame rate can be raised dramatically."

"Based on these figures, the dynamic range ought to be greater than 11 stops, and probably significantly better since in this test overexposure was not tested directly. The camera’s specification claims a signal to noise ratio of 66dB and 13.5 stops, but these noise measurements do not confirm that, 12 stops ought to be possible."

:sifone: Peter
 
You must log in or register to reply here.
Back
Top