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 :)

Epic-W Gemini Low-Light Test Footage + R3D

Adam Montville said:
A HUGE thing as well, that if the fill lighting is exactly the same, the Gemini isn't just 'cleaner' with noise structure, but it's also getting INFORMATION that's not there on helium and Dragon. Look at Dragon and Helium, and it's just black, with noise on top, where Gemini has clear shape and skin-tone coming out of that. Even at 6400, the mustache is a clear example of more detail, even though it's less resolution.

Seriously impressive.
Click to expand...
Yes I agree, there is much more information there. The difference is even bigger when you look at the JPGs on a big screen.
I read how the technique works (as far as we know) and I'm not surprised that it gives considerably better result. The signal is boosted so it moves away from the noise floor in the LL mode, but you loose dynamic range when you do this. In the normal mode, or the older sensors, the light we want is very close to the noise floor in strength so when we increase ISO in post we also lift up the noise floor which is what we see in his face on the dark side.

In fact the technique is so good so I would expect Red's other sensors to arrive with similar tech. However I can imagine that the technical challenges of doing the same for smaller photosites are considerable, so maybe it will take a while before it is possible.

I just looked at some of the Dragon and from what I can see 1600 ISO is border line usable, it is very similar to Gemini 12800. I would say Gemini is better at 12800 than Dragon at 1600, because the Dragon image is darker so there is less detail.

So from an ISO perspective maybe we could compare them like this
Dragon. 1600, maybe a bit less say 1200
Helium. 3200, maybe 4000
Gemini 12800

/Andreas
 
AndreasOberg said:
I did a crop of the jpegs at 6400 and 12800 ISO. I'm focusing mainly on high iso situation since for us this would open up a lot of situations.

41206341992_1180d72ba7_b.jpg

Hope you are ok with me uploading it Martin since it is of course your test :)

I recommend downloading all of the jpegs from Martin and go back and forward on a UHD TV. Our OLED TV show the differences in noise much clearer than our computer monitors.

From what I can see in this test, these are my conclusions:
- I would not use Helium higher than 3200 in this situation which is still impressive.
- Gemini seems usable all the way up to 12800 if you can do strong noise reduction, for example if the topic does not have a lot of detail.

If I compare Helium downsampled to UHD noise with Gemini I would say that Helium 3200 is a little bit cleaner than Gemini at 12800. That is indicating to me that downsampled to UHD Gemini is about 1 2/3 stop better than Helium. Possibly you could do more clever noise reduction on Helium since the resolution is higher.
If you would watch Helium at 100% and Gemini at 100% I would expect Gemini to be even better, maybe even more than 2 stop better.

Considering how good Helium already is when it is dark, Gemini is a big step forward in this regard.
Best wishes,
Andreas
Click to expand...

Looks good. I don't mind at all. So many ways to display the data, thanks for contributing.
 
So based on some feedback and questions I've amended the test video with an extra segment showing 12800 corrected using Resolves chart color match tool and applied medium denoise so that we can all see a potential more finished look as well as compare how well the 3 reds can match. The color match tool is auto based off the chart and I'm sure with a little more time and effort you could match the cameras even better. Here is a still grab for the moment until the video is updated.

Trio_12800_med_desnoise.jpg
 
Bob Gundu said:
Is that DSMC1 Dragon footage? My Weapon MG Dragon looks cleaner than this. Plus what OLPF?
Click to expand...
Maybe because Dragon looks more under exposed compared to Gemini. Just the LL Gemini advantage..
 
Bob Gundu said:
Is that DSMC1 Dragon footage? My Weapon MG Dragon looks cleaner than this. Plus what OLPF?
Click to expand...

Yes DSMC 1. Also are you comparing to yours based on new or old ISO calibration?
 
Jacek Zakowicz said:
Maybe because Dragon looks more under exposed compared to Gemini. Just the LL Gemini advantage..
Click to expand...

Lighting and settings were consistent. Also the "corrected" version as I mentioned is based on Resolves match feature. I went with this approach to give a technically matched image instead of based on my personal opinion. Skip to the end of the video to see this in motion, its now updated on vimeo.

Also here are some R3Ds to play with.
https://www.dropbox.com/sh/ol7k5e9z9qced0d/AAAmZahGe6-JxLeZ9budvbaea?dl=0
 
Last edited:
Martin Whittier said:
Yes DSMC 1. Also are you comparing to yours based on new or old ISO calibration?
Click to expand...

As Graeme explained to me, the circuitry in DSMC2 bodies create much cleaner images than DSMC1. Also, switching to LLO filter helps a lot too.
 
Bob Gundu said:
As Graeme explained to me, the circuitry in DSMC2 bodies create much cleaner images than DSMC1. Also, switching to LLO filter helps a lot too.
Click to expand...

I agree completely, but I feel introducing another OLPF opens up a whole different battery of tests. But you aren't wrong.
 
Standard and lowlight are very similar in exposure. I think standard OLPF makes lowlight obsolete.
 
I did not notice - but was any of the cross camera exposure footage shot with Gemini not in 'LL' mode? So 800 would be way overexposed in LL mode, right? And no ND were used in any of the tests? The bulb was on a dimmer to show saturation of highlights?

For Helium, I have noticed it gets dramatically cleaner when downsampled from 8k to 4k. To the point that you can't compare 8k to 5k and get same answer as 4k post of each.
Does the demo compare 1:1 with same framing? If so what was downsample algorithm?

Thanks for the timing of this test BTW.
 
Mike Finegan said:
I did not notice - but was any of the cross camera exposure footage shot with Gemini not in 'LL' mode? So 800 would be way overexposed in LL mode, right? And no ND were used in any of the tests? The bulb was on a dimmer to show saturation of highlights?

For Helium, I have noticed it gets dramatically cleaner when downsampled from 8k to 4k. To the point that you can't compare 8k to 5k and get same answer as 4k post of each.
Does the demo compare 1:1 with same framing? If so what was downsample algorithm?

Thanks for the timing of this test BTW.
Click to expand...

He answered several of your questions already.

Martin Whittier said:
800 ISO in standard mode, 1600-12800 in LL mode.
Click to expand...

Martin Whittier said:
400% is relative to the UHD timeline. I did this to help ensure viewers can judge the noise pattern regardless of online compression. All the reds were shot with the standard OLPF. 1.35 was the FLUT value to align the chart.
Click to expand...

Martin Whittier said:
SETUP
Camera w/ 50mm Ultra Prime (T Stop matching Key Light) @ 56000k
Key: Astra 1x1 w/ Snap Bag into a 4x4 Half soft Frost @ Minimal output - 5600K
Room Fill: Skypanel bounced off ceiling @ 1% - 5600K
Practical Bulb: 80% on Dimmer - 3200K
Click to expand...
 
Anton Shavlik said:
Standard and lowlight are very similar in exposure. I think standard OLPF makes lowlight obsolete.
Click to expand...

Is that true. I thought there would be more difference with the lowlight? Has anyone measure the difference in light?
Andreas
 
AndreasOberg said:
Is that true. I thought there would be more difference with the lowlight? Has anyone measure the difference in light?
Andreas
Click to expand...


I would have though the 'light' is the same but the noise characteristics and range would be different. So 800 at LL and Standard would look initially similar, except the LL would perhaps have less highlight. Just guessing based on what's been said though.

It would be a good test

cheers
Paul
 
paulcurtis said:
I would have though the 'light' is the same but the noise characteristics and range would be different. So 800 at LL and Standard would look initially similar, except the LL would perhaps have less highlight. Just guessing based on what's been said though.

It would be a good test

cheers
Paul
Click to expand...

One of Phil Holland's tests, Low Light OLPF is better than Standard in the low end noise, I think it was less than half a stop though. Then you also lose some of the color / flare fixes that the Standard brings as well, so LL should only be really used for exactly that, low light. At ISO 800, I would NOT use Low Light. At ISO 6400.... Yes, Low Light.

https://www.youtube.com/watch?v=CvtC1Kje4EQ
EDIT: Keep in mind that all of the ISOs in this video are the OLD cal.
 
paulcurtis said:
I would have though the 'light' is the same but the noise characteristics and range would be different. So 800 at LL and Standard would look initially similar, except the LL would perhaps have less highlight.
Click to expand...
Hiya Paul. Just to be clear I talk about the Low light OLPF (and not the LL mode in the camera). As you probably know the OLPF blocks or reflects certain amount of wave lengths to various degrees so you do actually have less light hitting the sensor.

/Andreas
 
You must log in or register to reply here.
Back
Top