GH5 to Alexa Conversion

[Attila Bakos 218]

13 hours ago, Sage said:

It must be pre-baked data sets of some sort ('assemble on site'). What Cpu are you rockin'?

It's an old i7-3770. One of the reasons I use Matlab is that many of it's commands are internally multi-threaded, it saves me a lot of time, especially with this old machine.

[Sage 682]

@Attila Bakos Very cool, I was on a 3770k; I switched to Ryzen last year for this very purpose. Now I'd like to swap out to one of the newest Ryzens, but they are sold out everywhere. I'm doing something atm wherein I throw enormous amounts of data in, and the renders can easily go upwards of an hour at 100% cpu. My data files are usually around 1Mb per each; these are going upwards of 50Mb atm

@zerocool22 CST is a versatile tool that allows whitepaper transforms between any two spaces. These are the manufacturer specified color spaces. Emotive is a measured conversion between two cameras (or profiles etc), such that it factors in sensor response to a light source and internal camera processing (of WB, profile, etc). Emotive has simulations of sensor LogC solely using Resolve color tools (sans CST) called Matrix. Both the interpolated conversions (with Tetrahedral interpolation enabled) and the Matrix LogC simulations are non-destructive, while Arri's Rec709 (that can be added to Matrix LogC) is partly destructive (of extreme Led gamut). CST can be either destructive or non-destructive, depending on the transform specified and relevant gamut.

CST + Arri 709:

Emotive Color

[HockeyFan12 441]

I usually work with Alexa footage that's shot at one of three presets (3200K, 4300K, 5600K). Sage’s LUTs work really well, but getting the best match requires white balancing both the S1 and Alexa to neutral white before converting. Alexa white balance presets are not neutral–they're orange/green.

I wanted equivalent white balance presets I could use (5600k, 4300K, 3200K) with the S1, where I could shoot V Log, apply Sage's LUT, and get a ballpark match to Alexa presets–even if it might be a little less precise than doing this the right way.

I lit a white card with a Luxli RGBW light at 3200K, 4300K, and 5600K and then shot Alexa footage shot at 3200K, 4300K, and 5600K presets.

Then I shot the same with an S1 and and applied Sage’s Log C conversion LUT. I adjusted the white balance settings manually until I got a good match.

I've only tried out the 5600K preset so far in the field but it matches pretty well.

K1: 5600K A: 6 G: 9 (should be G: 10, but it doesn't go that far) – apply Daylight Log C conversion 

K2: 4300K A: 6 G: 6 – apply 4200K Log C conversion

K3: 3200K: A7: G: 4 – apply Tungsten Log C conversion

Beyond that, a bit of extra noise and a bit of unsharp mask after downscaling S1 4k to Alexa 2k/3.2k (amount 24, radius 3) also helps make S1 footage look more like Alexa footage. 

If this isn’t for you, ignore this post. It's not as precise as doing things the right way but for my needs is much more convenient.

[TomTheDP 135]

Shot this with the S1 using emotive color LUT. 

[Sage 682]

On 12/15/2020 at 5:28 PM, TomTheDP said:

Shot this with the S1 using emotive color LUT. 

Lol @ the ending

[jack jin 16]

On 12/19/2020 at 10:25 PM, Sage said:

Lol @ the ending

Hi @Sage, I hope you've had a great new year so far, I have used the s1alex lut on various personal projects now, gotta say I love the colors that it gives me, much more consistent and pleasing results even in warm or weird lighting compared to the p4k version. But the exposure seems to be quite a bit brighter than the normal rec709 lut. And the lut shows a lot more noise and artifacts then I expected especially at the higher iso gain. I have tried to reduce the exposure using the exposure adjustment luts, but on darker shots it clips the shadows and doesn't roll them off naturally like the braw exposure adjustments. Any advice on how to reduce the noise and artifacts without using noise reduction? (PS: I am using tetrahedral 3d lut interpretation in Davinci Resolve already.)

[Sage 682]

Interesting; what comes to mind is that the latest files allow projection below zero. This only happens in theoretical space (i.e. below legal LogC / VLog), and projects sans clipping (an offset immediately after can bring sub-zero data into view).  This area shouldn't be encountered in real footage, so it may be the higher Isos which change the lower data range (S1a is optimized for VLog 640 Iso, 100 Cine-D, 400 HLG). Also, luma grade maneuvers in advance of the Pre/conversion is another way to bring this range into view.  I used to have an artificial rolloff to 0 black for theoretical black projection, but this caused problems downstream, so I removed that cap with the latest files. For noise, increased color separation will amplify noise, despite completely smooth tetra interpolation, if the footage has noise macroblocks. Higher bitrate codecs and lower Iso is optimal for clean imaging. Can you send me .tifs of the raw footage to noodle around with?