Helium weird bokeh / clipped highlights (RED tech please look at this)

[konrad grant]

If you want proper linear output to do these kinds of comparisons, export an EXR. The integer linear output is very conservative in how it tries to avoid producing clipped data.

Graeme

Graeme , my understanding is EXR supports 32bit integer , and 16bit half float types .. it doesn't clip..its floating point(ish)

so there is no linear "integer" output . ..and 16bit half float is very lossy ..clipped or unclipped.

i know this, because
a) ive written integer -> half float functions
b) if I convert r3d output from integers -> half floats , my inhouse compression systems ,that are already capable of producing lossless files half the size of j2k on a good day.. start doing it much more regular ..and significantly beats openEXR PIZ compression .. due to the highly wasteful/ lossy nature of the integer -> 16bit half float transformation . YUK!! .. of course , for the purpose you mention ie, testing linear output etc it may suffice ..

of course , if red ever decided to finally explain to us when our cameras are, or are not recording 12bit and 16bit raw data,..as ive politely asked for ..years.. i might consider openEXR for 12bit dervived source footage .. but alas ..

so since half floats are not suitable for 16bit integer sources IMO (unless you throw half your data away and can live with it..).. surely its better just to stay in the integer domain and clip when necessary later down the pipe for real post situations?

half float may handle the clipping better (it can after all produce files above and below traditional clip lines, so it doesn't clip ..) .. but it is not without cost .. as half floats can handle about 8192 shades at best (if normalized between -1 an 1..) .. 16bit integer is 65535 shades (just don't clip, and you're golden .. no wasteful sign bits and mantissa)

ps, redcine used to have linear as an option , but I haven't seen it there for years ? is it still there ? where did it go ?

 

[Björn Benckert]

Graeme , my understanding is EXR supports 32bit integer , and 16bit half float types .. it doesn't clip..its floating point(ish)

so there is no linear "integer" output . ..and 16bit half float is very lossy ..clipped or unclipped.

i know this, because
a) ive written integer -> half float functions
b) if I convert r3d output from integers -> half floats , my inhouse compression systems ,that are already capable of producing lossless files half the size of j2k on a good day.. start doing it much more regular ..and significantly beats openEXR PIZ compression .. due to the highly wasteful/ lossy nature of the integer -> 16bit half float transformation . YUK!! .. of course , for the purpose you mention ie, testing linear output etc it may suffice ..

of course , if red ever decided to finally explain to us when our cameras are, or are not recording 12bit and 16bit raw data,..as ive politely asked for ..years.. i might consider openEXR for 12bit dervived source footage .. but alas ..

so since half floats are not suitable for 16bit integer sources IMO (unless you throw half your data away and can live with it..).. surely its better just to stay in the integer domain and clip when necessary later down the pipe for real post situations?

half float may handle the clipping better (it can after all produce files above and below traditional clip lines, so it doesn't clip ..) .. but it is not without cost .. as half floats can handle about 8192 shades at best (if normalized between -1 an 1..) .. 16bit integer is 65535 shades (just don't clip, and you're golden .. no wasteful sign bits and mantissa)

ps, redcine used to have linear as an option , but I haven't seen it there for years ? is it still there ? where did it go ?

Linear is there. But you need to go preferences, advanced options. Restart and you find it with the others: redgamma etc.

 

[Graeme Nattress]

The REDCine-X display shows "linear" as the "gamma" and that is the integer linear I refer to. Please don't use it. It's conservatively scaled to avoid clipping and that is something you cannot account for in the math. Please use EXRs that have un-cliped and properly scaled float representations of the data.

Graeme

 

[paulcurtis]

Changing the ISO doesn't have any effect if gamma is linear. So these settings - Camera RGB and Linear gamma - are the closest to the RAW data.

If you export an EXR with two different ISO settings you will get two difference levels of exposure in that exr. However you can easily match the two clips by adjusting exposure in the exr after export. It appears nothing is clipped and all the data is there.

That would make sense to me.

But on those scopes the different ISOs would produce different waveforms and that was my point about the ISO for Dragon vs Helium being the same and that not being the right way to compare.

If Graeme also says the linear gamma setting in RedCine is being conservative then there could be other issues at play.

Where those R3Ds in that comparison screenshot available for download?

cheers
Paul

 

[Graeme Nattress]

If you export an EXR with two different ISO settings you will get two difference levels of exposure in that exr. However you can easily match the two clips by adjusting exposure in the exr after export. It appears nothing is clipped and all the data is there.

That would make sense to me.

But on those scopes the different ISOs would produce different waveforms and that was my point about the ISO for Dragon vs Helium being the same and that not being the right way to compare.

If Graeme also says the linear gamma setting in RedCine is being conservative then there could be other issues at play.

Where those R3Ds in that comparison screenshot available for download?

cheers
Paul

Yes, the linear mode in REDCine-X is conservative to avoid clipping. It is raw-ish data in a sense, but white balance is still going on. I (obviously) use the raw raw data. The clip point for Dragon and Helium is the same. Looking at raw data in this way doesn't really help you understand the D and H differences. They are different. Physics makes that inevitable. H is more sensitive, lower noise. D has larger pixel. D has a bit more highlight protection. H has a bit less highlight protection but more detail and lower shadow noise. To match highlight room on Dragon at ISO800 try Helium at ISO1280 - it should be a close enough match.

Graeme

 

[paulcurtis]

Yes, the linear mode in REDCine-X is conservative to avoid clipping. It is raw-ish data in a sense, but white balance is still going on. I (obviously) use the raw raw data. The clip point for Dragon and Helium is the same. Looking at raw data in this way doesn't really help you understand the D and H differences. They are different. Physics makes that inevitable. H is more sensitive, lower noise. D has larger pixel. D has a bit more highlight protection. H has a bit less highlight protection but more detail and lower shadow noise. To match highlight room on Dragon at ISO800 try Helium at ISO1280 - it should be a close enough match.

Graeme

I wasn't doing the comparing as i only have an E-W (which i think is awesome, and i love the shadows).

Just for my own sanity. When exporting as OpenEXR the gamma setting is basically ignored. But the colour does set the primaries accordingly, so RWG vs DragonColor looks different in the resulting EXR files.

So is there public details about the primaries for RWG yet, or will there be? Normally i would matrix the primaries myself in Nuke. Or is RWG not just a simple matrix operation and that's the whole colour science bit. And in order to move from RWG we should be looking towards non linear LUT transforms?

Not a big issue right now, more curiosity. Need to get the foundry to update the Red SDK being used first...

cheers
Paul

 

[Graeme Nattress]

Paul - drop me an email graeme@red.com

So yes, EXRs honour the colour space settings, so you can get an EXR out in RWG, or DC or DC2 or whatever.

Graeme

 

[Micah Simms]

Forgive me if this a completely noob question. Does the Epic-W ship with the REDWideGamutRGB? I am out of town and my Epic-w is packed up at the moment but I did check some footage on Red Cine X and didn't see the option to choose the REDWideGamutRGB. If it doesn't ship with it, can someone link me to a where I can install it?
Thanks!

 

[Graeme Nattress]

RWG is in post for the moment in the latest version of RC-X. We're working on bringing it to camera.

Graeme

 

[Micah Simms]

Great! Thanks Graeme!