Intermediate CODEC question

When the source material is 1080 25p 4:2:0 8bit h.264 and the delivery is the same, is there any benefit by using intermediate CODEC of 4:2:2 and/or 10bit for the sake of better color correction/grading?

Yes of course ! You can have different formats/containers/codecs/specs for capture, intermediate (post production) and final diffusion but you need to understand what your hardware and software stack is doing at every step and why you would need better quality.

For example if you do "8 bpc input -> multiple color grading in 8 bpc -> 8 bpc" output you can have weird artifact popping up and doing "8 bpc input -> multiple color grading in 16 bpc -> 8 bpc" instead can remove them. All depends on your actual content / editing / special effects of course.

Joshua is asking if there is any difference in having 8bit source converted to 10bit intermediate file vs. using original 8bit. Not a difference between 8 vs 10bit real source.
Answer is- no. Your 10bit intermediate file will hold same usable 8bit data. There is no magic (until you use some spacial tool which would try to increase bit depth, which is not easy).
Only difference is that intermediate codecs will make your editing smoother as they are way less compressed and easier to decode (and I frame based). When it comes to usable data for grading it won't give you anything above your original source.
10$ is 10$ regardless if you have it in big wallet or small one

Thanks, Antoine and Andrew.

Another consideration is the software you use for color correction: the internal operations need to happen in higher bit-depth if you don’t want to introduce any new artifacts especially with 8-bit footage during decode/encode. Many modern software like DaVinci will operate in 32-bit floating point, so you should not to worry working with lower bit footage in terms of image quality.


Sent from my iPad using Tapatalk

Thanks, David.
I use DaVinci Resolve for both editing and color correction/grading.

Joshua is asking if there is any difference in having 8bit source converted to 10bit intermediate file vs. using original 8bit. Not a difference between 8 vs 10bit real source.
Answer is- no.

The answer is "it depends" -- cf. David's answer and mine

Your 10bit intermediate file will hold same usable 8bit data. There is no magic

The magic comes from the fact that you loose quality each time you store the results of an operation on the data. If your logic was true then filmmakers could shoot in 8 bpc, edit and grade in 8bpc and have a great 8 bpc output...

Looks like you don't really understand original question. It's not about shooting 8bit vs 10bit (or editing/grading in higher than source bit depth), but about converting 8bit source into 10bit in hope for gaining usable data during this process.

Resolve will convert any source to 32bit float straight after decoding. So if you have 8bit file or previously converted 8bit to 10bit your usable data in Resolve will be the same and end results as well. Where does "your magic" happens here? Where will Resolve find more data in this 10bit source (coming from 8bit) file compared to original 8bit? In many cases it's same date with padding zeros (in others just converted 8bit into 10bit, which adds nothing new/usable). It's plain math. In order to get something more you need to have advanced algorithm added to bit depth conversion process, but not a single common tool does such a thing. There are some open source tools for this, but this is very specific topic.
10$ is 10$ regardless if it's in small or big wallet

Thanks, Andrew.

… In order to get something more you need to have advanced algorithm added to bit depth conversion process, but not a single common tool does such a thing. There are some open source tools for this, but this is very specific topic. …

What are those open source tools?

avisynth, vapoursynth with filters GradFun3, f3kdb 2.0, dither tools etc. Those filters can remove banding quite well in some cases (= increased bit depth), but when you encode to low bitrate then banding will be back. I don't think it's worth the hassle until you really have special case (eg 8bit source footage which needs to go to high quality end master).

Thanks, Andrew.

Looks like you don't really understand original question. It's not about shooting 8bit vs 10bit (or editing/grading in higher than source bit depth), but about converting 8bit source into 10bit in hope for gaining usable data during this process.

Resolve will convert any source to 32bit float straight after decoding. So if you have 8bit file or previously converted 8bit to 10bit your usable data in Resolve will be the same and end results as well. Where does "your magic" happens here? Where will Resolve find more data in this 10bit source (coming from 8bit) file compared to original 8bit? In many cases it's same date with padding zeros (in others just converted 8bit into 10bit, which adds nothing new/usable). It's plain math. In order to get something more you need to have advanced algorithm added to bit depth conversion process, but not a single common tool does such a thing. There are some open source tools for this, but this is very specific topic.
10$ is 10$ regardless if it's in small or big wallet

I hold what I'm saying. If one doesn't know the inner workings of the software, one cannot infer anything about the end quality. For example maybe converting to 10bpc forces the host software to do everything in 16bpc, hence better quality. Maybe not and it's a user option to force the host software to have high quality conversion / filters

If one doesn't know the inner workings of the software, one cannot infer anything about the end quality. For example maybe converting to 10bpc forces the host software to do everything in 16bpc, hence better quality. Maybe not

There is no "maybe." Resolve does everything in 32Float and input bit depth is irrelevant with regard to processing strategy.

Given Andrew's money analogy you're asking for a $7.49 bill which might exist in 10-bit and better, but does not in 8-bit. That's all there is to it. (I realized USA still has $1 bills... we don't in Canada, anymore, nor 1-cent "pennies".) If you want exact change, you're going to have to operate in a higher original bit-density or dither like hell.

There are cheats, but 8-bit is like hoping the black flies won't get into your tent when what you are using for a screen is a tennis net.

jPo, CSI

You're probably right if we restrict ourselves to DaVinci Resolve yes

I hold what I'm saying. If one doesn't know the inner workings of the software, one cannot infer anything about the end quality. For example maybe converting to 10bpc forces the host software to do everything in 16bpc, hence better quality. Maybe not and it's a user option to force the host software to have high quality conversion / filters

This is a very different matter. You worked too much with Premiere
If you don't tick Maximum Bit Depth Premiere will decode even 10bit codecs to 8bit pixel format for speed, no?
It all doesn't change fact that converting 8bit to 10bit doesn't add magically any more usable data.
It's all about knowing your app.

If you don't tick Maximum Bit Depth Premiere will decode even 10bit codecs to 8bit pixel format for speed, no?

In theory yes but in practice it depends on the actual Format/Codec chosen

It all doesn't change fact that converting 8bit to 10bit doesn't add magically any more usable data.
It's all about knowing your app.

Indeed! Also, there might be a "gain of data" if you're doing a RGB -> YUV transformation, I would need to study this more in depth

If you don't tick Maximum Bit Depth Premiere will decode even 10bit codecs to 8bit pixel format for speed, no?

In theory yes but in practice it depends on the actual Format/Codec chosen

It all doesn't change fact that converting 8bit to 10bit doesn't add magically any more usable data.
It's all about knowing your app.

Indeed! Also, there might be a "gain of data" if you're doing a RGB -> YUV transformation, I would need to study this more in depth

Interesting?
They come from where the 2 bits to go from 8 to 10 bits?

Interesting?
They come from where the 2 bits to go from 8 to 10 bits?

Salut Jean Claude,

It's not really that we "gain" 2 bits but more that we "loose" information by restricting the operations in a 8 bit space. Using bit depth as a measure for image quality is good -- up to a certain point. It's especially good to explain to the professional public. I'm sure better indicators are to be found in mathematics if ones really wants to maximize quality in its software.

If you want an example of what I'm talking about suppose you have a,b,c,d variables such as :

Code: Select all

a=0 ou 1 (1 bit)
b=0 ou 1 (1 bit)

c=a+b -> 0, 1 ou 2 (1.58 bits)
d=a-b -> -1, 0 ou 1 (1.58 bits)

To store "properly" (without loosing quality doing the reverse function) c and d you need more bits (4 bit because 2*log_2(3)=3.16 bits). But c and d "transport" only 2 bits of information (they are the result of a reversible function from a 2 bit space), the space of 4 bits will not always be used as its maximum as values are intertwined.

That's why I'm saying each step in a pipeline can lower quality and it's a huge mess. This was just an example, I'm not saying it actually happens in Premiere Pro or DaVinci Resolve

I guess my previous example answers Andrew's reformulation of the OP question

It's not about shooting 8bit vs 10bit, but about converting 8bit source into 10bit in hope for gaining usable data during this process

Answer is, it depends

I guess my previous example answers Andrew's reformulation of the OP question

It's not about shooting 8bit vs 10bit, but about converting 8bit source into 10bit in hope for gaining usable data during this process

Answer is, it depends

No, it doesn't depend. You're not going to gain anything by converting 8 bit source into 10 bit source before you feed it in into resolve. In fact, you actually lose quality because most codecs are lossy.

No, it doesn't depend. You're not going to gain anything by converting 8 bit source into 10 bit source before you feed it in into resolve. In fact, you actually lose quality because most codecs are lossy.

If you're unable to read previous comments and assess / criticize arguments we can't have a discussion and move forward. Yes, it does depend (and yes of course in some cases you can also loose quality in the process)

No, it doesn't depend. You're not going to gain anything by converting 8 bit source into 10 bit source before you feed it in into resolve. In fact, you actually lose quality because most codecs are lossy.

If you're unable to read previous comments and assess / criticize arguments we can't have a discussion and move forward. Yes, it does depend (and yes of course in some cases you can also loose quality in the process)

I can read just fine, you're the one that doesn't seem to be able to!

I'm a software developer by trade, and I went to college to be a physicist specializing in software and math. No simple transcode is going to make your footage better.

I'm a software developer by trade, and I went to college to be a physicist specializing in software and math. No simple transcode is going to make your footage better.

Having a science and engineering background is not a valid argument against my claims. And no transcode is ever "simple", you just don't know what happens behind the hood

I'm a software developer by trade, and I went to college to be a physicist specializing in software and math. No simple transcode is going to make your footage better.

Having a science and engineering background is not a valid argument against my claims. And no transcode is ever "simple", you just don't know what happens behind the hood

Yeah, okay I can see you're not worth wasting a single moment on!

There are cheats, but 8-bit is like hoping the black flies won't get into your tent when what you are using for a screen is a tennis net.

Sorry I missed this question a few months ago when it was first asked.

I agree with Joe: the way I would put it is, trying to get more out of converting 8-bit video to 10-bit is kind of like taking 8 gallons of water and pouring it into a 10-gallon drum. You don't really get any more out of it.

The right answer is not to shoot with 8-bit cameras, whenever possible. The more extreme the color-correction (or the processing), the more the image will tend to fall apart.

I am not a fan of 8-bit video, especially for post. I get that a lot of broadcast video winds up as 8-bit for delivery, though the hope is that 4K HDR (particularly Dolby Vision) allows for 10-bit or even 12-bit.

HDR is already driving the change, more and more cameras even in the lower segment are getting 10 bit.