Field comparison of New OLPF vs. Old OLPF (inspired by ISO vs. Film Grain)

[Michael Tiemann]

Reading Phil Holland's posts is a bit like listening to a Beatles record--you learn more and more as you experience more and more while reading the words over and over again.

I'm continuing my journey of learning the DRAGON, and I'd like to share a few recent learnings. Let me just say at the outset that nothing I plan to say should be read as a complaint. If I say "I could get X to do Y but not Z" that's not a complaint about X nor about Y. It's just data about where I am and what I've observed or figured out to this point.

I have an EPIC DRAGON with an old OLPF on long-term loan. I also have a SCARLET DRAGON with a new OLPF, fresh off the assembly line. Both are running the latest beta software, which means both have DRAGONColor and RG4 as baked-in look choices. I decided to do a somewhat unscientific test comparing the two OLPFs to see what I could learn. In truth, what I've done so far is to have confirmed pretty much what Phil has already presented, but to do so in the context of my shooting situation.

What I have seen is that the old OLPF is more tolerant of less light, meaning that chroma noise is not quite so prevalent when light levels start to drop, compared to the new OLPF. What I've also noticed (and expect will change with the new firmware/SDK that Jarred has recently mentioned) is that with the old OLPF, when there is chroma noise, it's pretty much blue, whereas with the new OLPF it's mostly red, with sprinkles of blue. Phil gives great examples here:

http://www.reduser.net/forum/showthread.php?117820-Red-Dragon-ISO-Texture-Versus-Film-Grain

What I have reproduced in my own environment (R3D files coming soon...they take a while to upload) pretty much shows (no surprise) that with lots of light at ISO 320, the images are creamy. With moderate light at ISO 800, the images just start to show grain. With marginal light at ISO 2000, the images have show plenty of grain. Not necessarily too much, but there's not question it's observable. Again, all of this can be seen (if not necessarily grokked the first time around) in Phil's post.

Now here's where I get into trouble. As I said above, this is an "unscientific" test, meaning that I'm not perfectly controlling the variables. One camera is an EPIC with old OLPF and a Fujinon Cabrio 19-90 and the other is a SCARLET with new OLPF and a Fujinon Premiere 24-180. It is possible that the Premier does render color more nicely than the Cabrio. Or it's possible that the new OLPF renders color better than the old OLPF. Or its possible that one lens + one OLPF is a marriage made in heaven whereas the other pairing is a marriage made in hell. Thus, I cannot definitively attribute to a single factor the superior color rendition of the SCARLET + new OLPF + Premier zoom vs. the EPIC + old OLPF + Cabrio zoom, but I definitely see it. And, quoting Phil (and Jarred) again, it could be that that little filtration of IR keeps the DRAGON sensor from confusing itself and picking bad colors when there are better colors to be found. In which case it would be the OLPF, not the lens, that really defines the difference.

In any case, I prefer the grain structure presented by the old OLPF, but I prefer the color quality of what I believe the new OLPF offers. The new OLPF also clearly has less flare and glare. It will be exciting to see what happens to the noise with the next round of firmware and software upgrades.

 

[Will Keir]

Looking forward to the new firmware too. Thanks for sharing.

It will be exciting to see what happens to the noise with the next round of firmware and software upgrades.

 

[Michael Tiemann]

OK, time for some updates. Here's the dropbox link to about 11GB of data: https://www.dropbox.com/sh/z87i4i2z83xqq61/AADkBL2MsLgq8zfvR2AtBv6ia

A004 and B004 are parallel directories of EPIC (A, old OLPF) and SCARLET (B, new OLPF). Each contain 5.5GB of data. Each clip is 24 frames shot at 4fps, 6K REDCODE 2:1, using approx 277MB per clip. There is a REDCINE-X project called AllClips-6KBS.rcx (the BS stands for Black Shade, because going from 1/4th second to 1/16th of a second without an intervening blackshade created a pretty large magenta color shift, even though there was no "E-" warning in the viewfinder).

There are three sources of lights:

(1) 300W of tungsten (halogen) overhead (always on full across a 12' wide, 4' deep console--think of it as 10W per sq ft from 5' overhead)
(2) a pair of Solectron LED lights that vary from 100% to 70, 50, 30, 20, 10, and 0% (balanced 50/50 warm/cool LEDS)
(3) a single ETC D22 LED light that also varies from 100% down to 0% as above (balanced to "WHITE", neither warm nor cool).

The sequence is in three cycles: 1/4 second at ISO 320, 1/16th second at ISO 800, 1/48th second at ISO 2000. The idea was to have some frames where the lighting clearly overwhelms the needs for the scene (clipping, perhaps) and otherwise needs to be toned down, all the way to where the lighting was clearly not enough for the scene (and noise creeps in). Obviously nobody would shoot video at 1/4th of a second...that's just an easy way to give me lots of apparent illumination without firing up the 2K. I intentionally shot straight into the Solectron LED panels to see what kinds of flare/glare the camers would experience. Normally I frame that shot so the lights are not visible. I wanted to see if I could get any magenta streaks that others have when shooting straight into lights. Alas, I could not. But the data's there, in case I'm not seeing it.

Part of the idea is to experiment with a range of light balances (tungsten and daylight-balanced sources) light intensities, and see what bends and what breaks under a variety of circumstances.

Also in the directory you will find an ADD TIFF from today's shots. I haven't decided whether to upload those yet or not, but instead of shooting wide, I shot a bit longer, cropping out the Solectrons (and not even using them). This gave the cameras a better chance to show off their ability to see fine details on can actually touch, rather than details that are small only because the shot is wide. As I mentioned on another post in this forum, there is definitely something sexy about the grain of ISO 2000 with the new OLPF. Not that that's the only way to go, but if you are going for organic grain, you don't need to buy a plug-in to get it ;-)

Here's a direct link to the ADD TIFF: https://www.dropbox.com/s/vw0ft1oaz6m9jfx/B005_C022_0623LF.13.0000013F.tif

Have fun!

 

[Phil Holland]

Nice Michael. Heh. Certainly the first time I've been compared to The Beatles I do choose my words carefully. Glad you picked up on that.

 

[Michael Tiemann]

Eye candy at ISO 320 and a decent amount of light from the SCARLET DRAGON with new OLPF:

https://www.dropbox.com/s/tzf3eloem5bdbqk/B005_C022_0623LF.0000013F.R3D
https://www.dropbox.com/s/oyw0hv4qg0xjk02/B005_C022_0623LF.0000013F.RMD

 

[Will Keir]

Beat me to it. Thanks Michael.

OK, time for some updates. Here's the dropbox link to about 11GB of data: https://www.dropbox.com/sh/z87i4i2z83xqq61/AADkBL2MsLgq8zfvR2AtBv6ia

A004 and B004 are parallel directories of EPIC (A, old OLPF) and SCARLET (B, new OLPF). Each contain 5.5GB of data. Each clip is 24 frames shot at 4fps, 6K REDCODE 2:1, using approx 277MB per clip. There is a REDCINE-X project called AllClips-6KBS.rcx (the BS stands for Black Shade, because going from 1/4th second to 1/16th of a second without an intervening blackshade created a pretty large magenta color shift, even though there was no "E-" warning in the viewfinder).

There are three sources of lights:

(1) 300W of tungsten (halogen) overhead (always on full across a 12' wide, 4' deep console--think of it as 10W per sq ft from 5' overhead)
(2) a pair of Solectron LED lights that vary from 100% to 70, 50, 30, 20, 10, and 0% (balanced 50/50 warm/cool LEDS)
(3) a single ETC D22 LED light that also varies from 100% down to 0% as above (balanced to "WHITE", neither warm nor cool).

The sequence is in three cycles: 1/4 second at ISO 320, 1/16th second at ISO 800, 1/48th second at ISO 2000. The idea was to have some frames where the lighting clearly overwhelms the needs for the scene (clipping, perhaps) and otherwise needs to be toned down, all the way to where the lighting was clearly not enough for the scene (and noise creeps in). Obviously nobody would shoot video at 1/4th of a second...that's just an easy way to give me lots of apparent illumination without firing up the 2K. I intentionally shot straight into the Solectron LED panels to see what kinds of flare/glare the camers would experience. Normally I frame that shot so the lights are not visible. I wanted to see if I could get any magenta streaks that others have when shooting straight into lights. Alas, I could not. But the data's there, in case I'm not seeing it.

Part of the idea is to experiment with a range of light balances (tungsten and daylight-balanced sources) light intensities, and see what bends and what breaks under a variety of circumstances.

Also in the directory you will find an ADD TIFF from today's shots. I haven't decided whether to upload those yet or not, but instead of shooting wide, I shot a bit longer, cropping out the Solectrons (and not even using them). This gave the cameras a better chance to show off their ability to see fine details on can actually touch, rather than details that are small only because the shot is wide. As I mentioned on another post in this forum, there is definitely something sexy about the grain of ISO 2000 with the new OLPF. Not that that's the only way to go, but if you are going for organic grain, you don't need to buy a plug-in to get it ;-)

Here's a direct link to the ADD TIFF: https://www.dropbox.com/s/vw0ft1oaz6m9jfx/B005_C022_0623LF.13.0000013F.tif

Have fun!

 

[Will Keir]

So that's how you get REDCODE 2:1. Shooting at lower FPS?

At 24 FPS, I only get REDCODE to 5:1, which looks great.
Did some tests tonight shooting 10:1 in some ambient darkness, we'll take a look at the noise level difference.

Each clip is 24 frames shot at 4fps, 6K REDCODE 2:1, using approx 277MB per clip.

 

[Timur Civan]

One thing to be careful of doing tests at long shutter speeds is that ambient light, in low light situations is often contaminated by low level reflections and is muddy by nature.

Try again with the 2k and I bet you will have a better time.

 

[Michael Tiemann]

One thing to be careful of doing tests at long shutter speeds is that ambient light, in low light situations is often contaminated by low level reflections and is muddy by nature.

Try again with the 2k and I bet you will have a better time.

I sense a mixing of apples and oranges here. I did not shoot a long shutter speed to get a low REDCODE...that was just coincidence. I shot a long shutter speed to get lots of light onto the sensor, and with the long shutter speed I could also shoot a low REDCODE.

Also, one of the things you will notice is that when the lights are up to 100%, they wash out the glow of the VU meters. When the lighting is down to 30% or 20%, the VU meters have a nice glow to them. When the lighting is at 0%, things do get muddy because there is not a well-defined lighting source providing an overall sense of direction. I shot at all these different levels of lighting precisely so that I could see what parts of the image break down at various points...whether it's the lighting that's falling apart, the sensor, both, or neither.

 

[Antoine Baumann]

In any case, I prefer the grain structure presented by the old OLPF, but I prefer the color quality of what I believe the new OLPF offers. The new OLPF also clearly has less flare and glare. It will be exciting to see what happens to the noise with the next round of firmware and software upgrades.

This is also what I think. Really hoping we can get both