F5/F55 Sensor resolution and bayer pattern

[Admit One]

Hallelujah.




LOG is a gamma curve which increases the "captured" dynamic range by compressing the highlights. It's characteristic is that it exposes much more similar to film when it comes to highlights which tend to have smoother highlights. LOG can be over-exposed slightly, but it is usually best to underexpose is because the highend is so heavily compressed and has much less data available in it for grading and pulling detail back out. The downside of LOG is you really want to shoot it in the highest bit-depth possible because each stop is not getting a uniform amount of data for it ie. remember it is a curve that flattens out near the top in the highlights which as mentioned are "compressed" and close together.

With linear RAW however each stop of exposure would get the same amount of data giving you a lot of flexibility in post for grading any aspect of the full tonal range. ie. With 16-bit-depth you get approx. 65,000 steps ÷ 14 stops = 4642 steps per stop of data BUT EQUAL THROUGHOUT THE ENTIRE RANGE of your image.

To me, in my opinion RAW is way more flexible to shoot in terms of exposure whereas with SLOG I you need to nail it within a stop and use extremely high bit-rates to get the most out of it. Even 10-bit SLOG which the F3 lets you do is not enough in my opinion as it only yields 73 steps per stop on average (again not equal per stop since its a curve).

While the cameras also output other formats, I was referring to their RAW output.

The Alexa and Blackmagic cameras output 12 bit log, and it is RAW. The C500 outputs 10 bit log and it also is RAW. The Sony's output 16 bit linear RAW.

Cheers.

 

[nyvz]

Hallelujah.




LOG is a gamma curve which increases the "captured" dynamic range by compressing the highlights. It's characteristic is that it exposes much more similar to film when it comes to highlights which tend to have smoother highlights. LOG can be over-exposed slightly, but it is usually best to underexpose is because the highend is so heavily compressed and has much less data available in it for grading and pulling detail back out. The downside of LOG is you really want to shoot it in the highest bit-depth possible because each stop is not getting a uniform amount of data for it ie. remember it is a curve that flattens out near the top in the highlights which as mentioned are "compressed" and close together.

With linear RAW however each stop of exposure would get the same amount of data giving you a lot of flexibility in post for grading any aspect of the full tonal range. ie. With 16-bit-depth you get approx. 65,000 steps ÷ 14 stops = 4642 steps per stop of data BUT EQUAL THROUGHOUT THE ENTIRE RANGE of your image.

To me, in my opinion RAW is way more flexible to shoot in terms of exposure whereas with SLOG I you need to nail it within a stop and use extremely high bit-rates to get the most out of it. Even 10-bit SLOG which the F3 lets you do is not enough in my opinion as it only yields 73 steps per stop on average (again not equal per stop since its a curve).

Isn't it sort of the other way around? When talking about how RAW is stored, linear is essentially linear-light straight off the sensor/ADC with only minor corrections, right? Not something stored gamma-corrected. In which case, the brightest stop is represented by about 32,000 steps, second stop about 16,000, and the 12th stop is represented by only about 16 steps. But then again, since the 16bit ADCs receives sensor voltage directly and linearly, you'll never get more steps than that anyway unless you interpolate values, but the shadows most susceptible are almost always naturally "dithered" by noise and not really in need of more steps.

My understanding is that log encoding is designed to evenly space out those stops completely evenly within a bit-limited signal. It remedies the fact that light is logarithmic but sensors and ADCs are linear-light. So for example the same 16-bit linear RAW data could be converted and stored in 12bit log and each stop might get 256 steps, a great reduction of steps for the bright stops and interpolation for the shadow stops, but still plenty of data for any natural scene.

Log doesn't have as drastic a difference from gamma-encoding, since 2.2 gamma is closer to log than linear light, but gamma encoding is not used for RAW storage, only for debayered YUV/RGB.

I think the comparison gets confusing since in the YUV/RGB world, things are almost never encoded linear-light because it is a waste of space, so linear usually refers to gamma-encoded. But that is just my understanding, someone correct me if I'm wrong.

One big question for me will be how the Sony's 16bit linear version of RAW compares with log versions.

Both Alexa and Blackmagic are using 12bit log, and the C500 is using 10bit log.

Cheers.

I would not worry about that since at best 16bit linear is the closest thing to what is coming right off the ADC (which is essentially always linear itself and not likely more than 16bits) so it will not likely be an issue for quality. I think it will be a while before cameras are creating images for which a 16bit linear or 12bit log RAW container will be a limitation. It might not be far off though, since it may happen when cameras reach 15-16stops of DR and things wont fit in a 16bit linear container anymore.

It makes sense that Alexa and BMC are using log, since as I understand it they are dual read-out cameras, so they probably have a lot of data (two frame readouts) to combine and pack down into their RAW containers. Perhaps once the two readouts are combined they have extra data that might not quite fit as 16bit linear. Not to say it wouldn't fit in 16bit linear but it wouldn't be a good fit since it is processed/combined data, not linear sensor data. But I'm kind of just guessing here.

 

[starcentral]

 

[Kolor-Pikker]

Isn't it sort of the other way around? When talking about how RAW is stored, linear is essentially linear-light straight off the sensor/ADC with only minor corrections, right? Not something stored gamma-corrected. In which case, the brightest stop is represented by about 32,000 steps, second stop about 16,000, and the 12th stop is represented by only about 16 steps. But then again, since the 16bit ADCs receives sensor voltage directly and linearly, you'll never get more steps than that anyway unless you interpolate values, but the shadows most susceptible are almost always naturally "dithered" by noise and not really in need of more steps.

My understanding is that log encoding is designed to evenly space out those stops completely evenly within a bit-limited signal. It remedies the fact that light is logarithmic but sensors and ADCs are linear-light. So for example the same 16-bit linear RAW data could be converted and stored in 12bit log and each stop might get 256 steps, a great reduction of steps for the bright stops and interpolation for the shadow stops, but still plenty of data for any natural scene.

Yes, digital sensors are by nature linear, and linear raw is a direct representation of what the sensor captures; the brightest stop encompasses half of all possible shades, and each stop below that has half as many as the one above.
A gamma curve is usually applied by the camera's processing to make the linear image appear natural to our eyes, which are log sensors, and as such treat each group of tones as having an equal number of shades - this means compressing the highlights and expanding the mid-tones.
And yes, log is indeed a gamma curve designed to maximize dynamic range within an encoded (non-raw) signal.

 

[Admit One]

And yes, log is indeed a gamma curve designed to maximize dynamic range within an encoded (non-raw) signal.

Yes. And RAW log is not S-log, although both are also designed to allocate bits closer to our logrithmic way of seeing.

Arri, Canon and Blackmagic have choosen RAW log, while Sony has chosen linear RAW. But Sony is compressing 3.6/1.

Both methods involve discarding data. It remains to be seen whether or not there is any substantial difference.

Cheers.