10 Bit vs 8 Bit - Sony FS5 vs Sony a7Sii a7Rii - Is 10 Bit Always Better?

[kidzrevil]

10bits of data from a shitty sensor is still 10bits of data from a shitty sensor. None of it really matters if you expose well and your grading doesn't require you to make crazy adjustments to the image. Not even raw can save you from a crap exposure etc.

imo color science and dr matters more. I had such high expectations from my gh4's 10bit output and I was disappointed in the fact that it looked too close to the original. Made little to no difference in image quality vs 8bit.

now I wait patiently for a Canon 5d3 raw or blackmagic user to tell me my statement is bullshit and I need to have a seat somewhere LOL

[tupp]

19 hours ago, jcs said:

Hey tupp, 420 is full Y (Luma) and 1/2 resolution (both vertical and horizontal (could say 1/4 for both)

Where did 420 come from?  We are talking about 422.    You started with:

On 5/12/2016 at 7:21 PM, jcs said:

4K 8-bit 422 becomes pseudo 10-bit Luma 8-bit Chroma 444 1080p. 'pseudo' as the 4 8-bit sample's variation summed to 10-bit is helped by noise/dither.

 

 

 

19 hours ago, jcs said:

If we downsample 4K 420, we'll average together 4 Y's to get one new, low-pass-filtered (noise and alias reduced) pixel.

Don't average! -- NEVER AVERAGE!!    Always SUM!

 

Don't sacrifice overall accuracy/depth for a few wayward pixels.  Reduce noise some other, more direct way.

If Dugdale started out with 420 instead of 422, of course, that affects the end result.  However, my point in response to your hypothetical color depth equivalence example is that the color depth is essentially identical in these three scenarios:
4k-UHD, 422, 8-bit;
full-HD, 444, 10-bit-luma/8-bit-chromas (your example);
full-HD, 422, 10-bit.

[jcs]

1 hour ago, tupp said:

Where did 420 come from?  We are talking about 422.    You started with:

 

 

 

Don't average! -- NEVER AVERAGE!!    Always SUM!

 

Don't sacrifice overall accuracy/depth for a few wayward pixels.  Reduce noise some other, more direct way.

If Dugdale started out with 420 instead of 422, of course, that affects the end result.  However, my point in response to your hypothetical color depth equivalence example is that the color depth is essentially identical in these three scenarios:
4k-UHD, 422, 8-bit;
full-HD, 444, 10-bit-luma/8-bit-chromas (your example);
full-HD, 422, 10-bit.

A7x II is 4K 420- 422 is my typo. The samples are averaged in floating-point- after summing we multiply by .25: 1.0 is max value, the fraction contains the extra info. I'm a developer- this is how we write the code.

You examples for color depth equivalence is likely true in practice- folks would not be able to tell the difference (I think you meant 4K 420 8-bit, which converts to 1080p as 444 ~9-bit).

[tupp]

5 hours ago, jcs said:

 after summing we multiply by .25: 1.0 is max value, the fraction contains the extra info. I'm a developer- this is how we write the code.

No.  Don't do that

 

By summing four values and then multiplying by 0.25, you are actually averaging -- you are not summing.  The best way to convert from 4k, 8-bit to HD, 10-bit is to simply sum the four values, which retains the full color depth/accuracy of the original image and which is a perfect mathematical conversion between 8-bit and 10-bit.  No multiplier is necessary after summing.

 

5 hours ago, jcs said:

You examples for color depth equivalence is likely true in practice- folks would not be able to tell the difference (I think you meant 4K 420 8-bit, which converts to 1080p as 444 ~9-bit).

Your original hypothetical scenario has equivalent color depth to 1080p, 444, "~9-bit."  If you start with 420, the color depth would be less than that original scenario.