How did Claudio Miranda's test go?

[Kevin Lang]

Very nice work David! I see you worked with Dan Myrick he is a old friend of mine from the Blair Witch days.

 

[Evin Grant]

Everything else being equal I think it's worth noting just how similar the images in Claudio's test are. Yes there are differences in DR and some other small issues but I'm still flabbergasted by how similar this $25,000 camera (Kit price) is to this $250,000 camera. Including shadow noise level and of course detail, which is actually better on the Red along with real 35mm DOF.

 

[Finner]

As soon as you change your avatar...

Jim

At your wish sir.

Lost the leather jacket (black leather just is not as hip as it use to be). Also went with a seasonal flair. Hope you enjoy it.

Now about those questions.

 

[Shawn Nelson]

Lol, Jim and his penchant for ambiguous elipses...

 

[Jannard]

I think it is time for me to go offline for a bit...

Jim

 

[Finner]

Originally Posted by Jannard
That was clever... I guess my best response to your new avatar is... "go read my old posts on this subject". All your questions have already been answered.

Jim

Which posts of your 1,229 would you suggest?

Come on Jim cut me a break, the 3 would be quick to answer.

 

[Jannard]

1. Fixed with OLPF change
2. Fixed
3. Don't know what he is looking at since I haven't seen the footage.

Now I'm going offline. Happy holidays to all.

Jim

 

[Finner]

Thanks Jim.

When the Hoff comes to get #993 I have a big black speedo hug for ya.

Big happy holidays right back at ya.

 

[Warren Kommers]

Although RED has a huge price advantage, 35mm DOF(which is huge), and more resolution, the F23 clearly has some things going for it at this point. Fortunately I do work on productions that can afford F23's and 35mm so the price thing isn't the biggest thing for me. I guess I never really understood how beta RED really was. It's cool that it was tested in the public this way and also very productive in terms of the feedback it generated for RED. This may end up being the biggest assest the camera has. The community.

However I hope this new camera can ball with that 11 stops the f23 has on Figueroa in Claudio's test. It looks good! I'm rooting for RED all the way though.

Although I'm curious what the threshold is for dynamic range, noise, and other image charcteristics until it just comes down to what's inherent in the sensor.

Also is it impossible to make the rolling shutter edge softer? Like a soft wipe in a transition? I'm sure I'm not the first to suggest that.

 

[Philip Powell]

Noise comment

Noise comment

I really enjoyed reading Claudio's conclusions. The only thing that surpised me a bit was his note on noise.

Under advantages of the F23 he wrote
"• Noise is far less and when you do see it it looks much more pleasing."

This kind of shocks me, considering how the Red is often sited as being so low noise Plus, several have suggested how easy it is to pull a key with as it's so low noise. I expected quite a bit of what he had to say, with dynamic range etc., but not the noise comment.

 

[Warren Kommers]

"• Noise is far less and when you do see it it looks much more pleasing."

This surprised me as well. Especially follwing the general rule of thumb with sensor size and relative noise size of 4k to 1080.

 

[Evin Grant]

The second to last two frames of the test really show the difference in noise characteristics. Personally I don't see the f23 being cleaner, just having a tighter grain/noise structure. The Red does display a bit more random blue channel noise but this may be fixable in processing. I've seen similar low light footage that's cleaner before.

 

[Gavin Greenwalt]

I've noticed the red channel clip 'early' in lots and lots of shots. I think it just goes with the territory of having a green biased sensor. Often the data is still 'discoverable' in the red channel but you have to throw the white balance completely out of wack to find it.

If you have to boost the red channel by a stop to balance the light loss at the filter you're losing a stop of latitude.

(The biggest reason why I want a REDStream color mode)

I asked Mr Miranda specifically about this in an email and he said that the red channel was clipped in the raw data and was unrecoverable in the tail lights. So while it is something you have to be doubly careful of when checking your exposure (remember you lose a stop of highlight information in the reds under many lighting temperatures after WB) this was not a case where white balance was causing the clip.

 

[Adrian T.]

I asked Mr Miranda specifically about this in an email and he said that the red channel was clipped in the raw data and was unrecoverable in the tail lights. So while it is something you have to be doubly careful of when checking your exposure (remember you lose a stop of highlight information in the reds under many lighting temperatures) this was not a case where white balance was causing the clip.

That's exactly why we need histogram and false color indicators based on RAW data instead of RGB preview data. Then we could detect such a clipping quite easily.

 

[David Mullen ASC]

But what good would that do? You don't expose a night exterior scene for the bright red tail lights in the shot (unless they are big in frame), you expose to get max information all around. Underexposing another stop or so would have ruined the night shots; I'd rather live with the red lights clipping in that case.

 

[I Bloom]

I really enjoyed reading Claudio's conclusions. The only thing that surpised me a bit was his note on noise.

Under advantages of the F23 he wrote
"• Noise is far less and when you do see it it looks much more pleasing."

This kind of shocks me, considering how the Red is often sited as being so low noise Plus, several have suggested how easy it is to pull a key with as it's so low noise. I expected quite a bit of what he had to say, with dynamic range etc., but not the noise comment.

12:1 compression of noise does not look so good, since truly random noise is mathematically impossible to compress losslessly.

Hear is what a piece of Redcode looks like from a capped body:
How I created this image: I put the body cap on Red #87, (FYI this camera has not been updated) rolled for a few seconds.
Opened the R3D in RedAlert Exported 4K Tiff:
ISO 320,Kelvin 5600, Tint 0, Rec709, Exposure -.4, Saturation 1, Contrast 0, Brightness 0, RGB 1, Curve Flat
Opened Tiff in photoshop,
Applied levels 12, 1.0, 19
Cropped off a random little piece, scaled 400% with nearest neighbor, exported jpeg.

As a visual reference I created an almost black 16 bit file in Photoshop and used the gaussian noise filter to create a similar looking histogram, then I applied the same levels and scaling.

Okay, so what are we looking at, well noise on the Mysterium is going through transformations, first its being compressed, and then it's being debayered. So it doesn't look like the straight up 4:4:4 RGB noise you are expecting, it looks kind of clumpy. I think this is what Claudio is refering to when he says it's "not as pleasing".

Another thing to consider is that there is no signal here just noise. I think if you give Redcode an underlying signal it might yeild alot more pleasing results. The unpleasing noise appears to only be an issue in night exteriors where the image must be pushed hard and their is a ton of negative space.

Everything is a trade off. The F23 versus Red test is not totally fair since Red is compressing 12:1 plus bayer pattern and the F23 is not (Am I right?). A better test would be Red vs. Cineform compressed F23 footage. Or F23 versus uncompressed Red from the future optical Raw port.

I think it's clear that the noise level on both cameras is in fact extremely low.

IBloom

 

[Jannard]

We still have some tricks up our sleeves... remember that we are new at this stuff. To be compared favorably, at this point in our development, with the F23 is a good thing. These tests also show us that we need to continue to improve, which is our intention. The F23 has a 2/3" sensor and uses B4 lenses. That can't change (until Sony releases the F35 for $350,000). Our image can change and will continually get better. The "Christmas Cameras" prove that. And that isn't the end of it.

We consider the F23 and Genesis to be the primary digital competition at this point. If we are in the game this early on, and at $17,500... we are doing our job. But it isn't done yet.

Jim

 

[Warren Kommers]

But what good would that do? You don't expose a night exterior scene for the bright red tail lights in the shot (unless they are big in frame), you expose to get max information all around. Underexposing another stop or so would have ruined the night shots; I'd rather live with the red lights clipping in that case.

Mos Def David. Its trouble in film as well. However in a film DI you can select that strong red color and bring it down individually because there is information still there.

Sometimes Ive been lighting lately with intentionally stronger colors so they are easier to control individually in CC.

 

[Warren Kommers]

It is heavy competition Jim and you guys are looking good in many areas. It does seem like there should be more to be dug out of that sensor.

 

[I Bloom]

One thing I'd contend with Claudio's conclusion is that he puts the 12:1 compression in the disadvantages column. For everyone but the top tier, that can handle uncompressed 2K, that belongs in the other catagory.

IMO, 12:1 onboard compression is Red's main advantage.

IBloom