How Important is 10-Bit Really?

[Emanuel]

9 minutes ago, Deadcode said:

On your sample i can see banding on the 8 bit footage, but the bitrate is very low (16Mbps).

Low bit rate will invariably offer banding.

[cantsin]

Okay, now a lab experiment with the highest-possible quality 8bit 4:2:0 material that can be generated:

1. Converted the 16bit TIFFs into 8bit 4:2:0 JPEGs at 100% compression quality (using Imagemagick with the settings "-depth 8 -sampling-factor 4:2:0 -quality 100").
2. Imported the JPEG sequence into Resolve, applied color space transform Vlog -> Rec709
3. Used the "Grab still" function and exported the still frame as a PNG.

Here's the result:

[cantsin]

And finally, the very best-possible 8bit quality, 8bit PNG (RGB, equivalent to 4:2:2 YUV, lossless compression) derived from uncompressed 16bit TIFF, converted from VLog to Rec709. With an 8bit video camera, you would only achieve this quality if you externally recorded an uncompressed 8bit 4:2:2 HDMI signal to a lossless codec:

[Attila Bakos]

Interesting, 4:2:2 shows almost no difference compared to 4:2:0. Bandwidth does matter, but 10bit has a clear advantage.

[anonim]

18 minutes ago, cantsin said:

And finally..

Great effort. Thank you for dedication of your time - indeed, such a nice example of behaving...

[cantsin]

23 minutes ago, Attila Bakos said:

Interesting, 4:2:2 shows almost no difference compared to 4:2:0. Bandwidth does matter, but 10bit has a clear advantage.

...yes, even when it's highly compressed. That somewhat surprised me, too. 

One should not forget though that all my tests are specific to Panasonic's VLog curve - which highly compresses dynamic range and obviously has not been engineered for 8bit cameras. I probably would have got different/better 8bit results if I had, for example, applied a Canon C-Log curve (which was originally engineered for the 8bit 4:2:2 signal of the first-generation Canon C300 and C100).

[tupp]

3 hours ago, cantsin said:

The results are attached here. "8bit" and "10bit" only refer to the color depth of the original video file; both stills are 8bit PNGs.

Actually,  "8bit" and "10bit" refer only to bit depth -- bit depth and color depth are two different properties.

26 minutes ago, cantsin said:

One should not forget though that all my tests are specific to Panasonic's VLog curve - which highly compresses dynamic range

My understanding is that Vlog doesn't actually change the dynamic range -- it changes the tone mapping within the dynamic range to capture more shades in desired exposure regions.

[hijodeibn]

24 minutes ago, cantsin said:

...yes, even when it's highly compressed. That somewhat surprised me, too. 

One should not forget though that all my tests are specific to Panasonic's VLog curve - which highly compresses dynamic range and obviously has not been engineered for 8bit cameras. I probably would have got different/better 8bit results if I had, for example, applied a Canon C-Log curve (which was originally engineered for the 8bit 4:2:2 signal of the first-generation Canon C300 and C100).

Thanks for your effort cantsin....very interesting your results.....I would love to see something similar for Canon C-log....8-bit 420 and 8-bit 422 are almost the same in your results and in canon C-log I am expecting a bigger difference...

[cantsin]

22 minutes ago, tupp said:

Actually,  "8bit" and "10bit" refer only to bit depth -- bit depth and color depth are two different properties.

My understanding is that Vlog doesn't actually change the dynamic range -- it changes the tone mapping within the dynamic range to capture more shades in desired exposure regions.

"Color depth or colour depth (see spelling differences), also known as bit depth, is either the number of bits used to indicate the color of a single pixel, in a bitmapped image or video frame buffer". -  https://en.wikipedia.org/wiki/Color_depth

I didn't write that Vlog changes the dynamic range, only that it compresses it into a flatter curve.

- Can't test CLog with Resolve btw. because it's not included in the Color Space Transform tool. (And it's no longer really in use anyway, having been superseded by Canon Log 1/2.)

[HockeyFan12]

1 hour ago, cantsin said:

And finally, the very best-possible 8bit quality, 8bit PNG (RGB, equivalent to 4:2:2 YUV, lossless compression) derived from uncompressed 16bit TIFF, converted from VLog to Rec709. With an 8bit video camera, you would only achieve this quality if you externally recorded an uncompressed 8bit 4:2:2 HDMI signal to a lossless codec:

Interesting. I've done similar tests with noisier cameras (C300, Alexa, F35) and not seen any banding because of the heavy dithering from the noise in the source footage. 

I suppose it goes to show it's worth doing your own tests!

Certainly it seems to make a big difference here.

[hijodeibn]

40 minutes ago, HockeyFan12 said:

Interesting. I've done similar tests with noisier cameras (C300, Alexa, F35) and not seen any banding because of the heavy dithering from the noise in the source footage. 

I suppose it goes to show it's worth doing your own tests!

Certainly it seems to make a big difference here.

I just took the C100 and shoot the sun with internal recording 8-bit 420 and external recording 8-bit 422, then in resolve I took the same frame for both recording and exported as TIFF, I see some difference but minimal, and I don't see the extreme banding showed before.... @cantsin probably you could play with the files and see how bad is really the banding in C-log...

8-bit 420

c-log 420_1.1.1.tif

8-bit 422

c-log 422_1.1.2.tif

 

[Mark Romero 2]

57 minutes ago, hijodeibn said:

I just took the C100 and shoot the sun with internal recording 8-bit 420 and external recording 8-bit 422, then in resolve I took the same frame for both recording and exported as TIFF, I see some difference but minimal, and I don't see the extreme banding showed before.... @cantsin probably you could play with the files and see how bad is really the banding in C-log...

8-bit 420

c-log 420_1.1.1.tif

8-bit 422

c-log 422_1.1.2.tif

 

I'm sure it's there, but I have a hard time seeing the differences exactly. Maybe someone could point them out to me?

Also, what is the bit rates of the C100 when recording internally 420 and recording externally 422? Guess I could look them up but too lazy 

[hijodeibn]

2 hours ago, Mark Romero 2 said:

I'm sure it's there, but I have a hard time seeing the differences exactly. Maybe someone could point them out to me?

Also, what is the bit rates of the C100 when recording internally 420 and recording externally 422? Guess I could look them up but too lazy 

C100 internally is recording 8-bit 420, and externally is 8-bit 422

[tupp]

6 hours ago, cantsin said:

"Color depth or colour depth (see spelling differences), also known as bit depth, is either the number of bits used to indicate the color of a single pixel, in a bitmapped image or video frame buffer". -  https://en.wikipedia.org/wiki/Color_depth

The notion that bit depth is identical to color depth is a common misconception, that has apparently made it's way into Wikipedia.

 

The only instances in which bit depth and color depth are the same is when considering only a single photosite/pixel or a single RGB pixle-group.  Once extra pixels/pixel-groups are added, color depth and bit depth become different properties.  This happens due to the fact that resolution and bit depth are equally weighted factors of color depth, in digital imaging.

[cantsin]

9 minutes ago, tupp said:

The notion that bit depth is identical to color depth is a common misconception, that has apparently made it's way into Wikipedia.

 

The only instances in which bit depth and color depth are the same is when considering only a single photosite/pixel or a single RGB pixle-group.  Once extra pixels/pixel-groups are added, color depth and bit depth become different properties.  This happens due to the fact that resolution and bit depth are equally weighted factors of color depth, in digital imaging.

Do you have any reference for this? I couldn't find a single one online.

5 hours ago, hijodeibn said:

I just took the C100 and shoot the sun with internal recording 8-bit 420 and external recording 8-bit 422, then in resolve I took the same frame for both recording and exported as TIFF, I see some difference but minimal, and I don't see the extreme banding showed before.... @cantsin probably you could play with the files and see how bad is really the banding in C-log...

 

I can't because Resolve doesn't have built-in C-Log transformations, sorry.

[Mark Romero 2]

3 hours ago, hijodeibn said:

C100 internally is recording 8-bit 420, and externally is 8-bit 422

Oh... what I meant was, how many Mbs is the codec? Is the external 8-bit 422 ProRes?

Also, was hoping someone could point out the OPTICAL differences in those two tiff files. I downloaded them and looked at them and couldn't really see a difference. Maybe someone can help me look at what part of the photos that the optical differences are most apparent.

[User]

6 hours ago, Mark Romero 2 said:

Also, was hoping someone could point out the OPTICAL differences in those two tiff files. I downloaded them and looked at them and couldn't really see a difference. Maybe someone can help me look at what part of the photos that the optical differences are most apparent.

I don't know if it helps but, as I understand it based on test from at least 2014, that there is almost no perceived difference if using an external recorder. It's infinitesimal. 

The right 8 bits.

[hijodeibn]

8 hours ago, Mark Romero 2 said:

Oh... what I meant was, how many Mbs is the codec? Is the external 8-bit 422 ProRes?

Also, was hoping someone could point out the OPTICAL differences in those two tiff files. I downloaded them and looked at them and couldn't really see a difference. Maybe someone can help me look at what part of the photos that the optical differences are most apparent.

Internal recording is AVCHD 24 MB, external recoding is HQ ProRes 422.....

[tupp]

18 hours ago, cantsin said:

Do you have any reference for this? I couldn't find a single one online.

I don't have a reference.  When I studied photography long before digital imaging existed, I learned that resolution is integral to color depth.

 

Color depth in digital imaging is exceedingly more quantifiable, as it involves a given number of pixels with a given bit depth, rather than indeterminate dyes and grain found in film emulsion (and rather than unfixed, non-incremental values inherent in analog video).  The formula for "absolute" color depth in RGB digital imaging is: 

Color Depth = (Bit Depth x Resolution)^3

[cantsin]

4 hours ago, tupp said:

Color Depth = (Bit Depth x Resolution)^3

I found exactly three references for this equation, all in camera forums, and all posted by a forum member called tupp...

But seriously, I see the point that in analog film photography with its non-discrete color values, color depth can only be determined when measuring the color of each part of the image. Naturally, the number of different color values (and thus color depth) will increase with the resolution of the film or the print.

In digital photography and video, however, the number of possible color values is predetermined through the color matrix for each pixel. Therefore, in digital imaging, color depth = bit depth.