Difference between 4.2.2 and 4.4.4 for a VFX heavy movie

X

xhridhar

Active member
Hello,

We are planning on buying a F3 for shooting a movie which is very VFX heavy. I have been reading all the threads for a few weeks now but it is not very clear to me if one really needs 4.4.4 and S-log etc for grading and keying?

Is the difference very significant between 4.2.2 and 4.4.4? Given that one has to pay almost twice the price of the camera (firmware upgrade + Gemini + SSDs) for the 4.4.4 option and now that one has cheaper options like the upcoming Blackmagic product, is 4.2.2 more sensible option to take?

We have an in-house VFX artists team and so if 4.2.2 takes a bit longer, it is acceptable.

Nate weaver and Timur Civan seem to think that 10 bit is a bigger deal than 4.4.4 and that is covered by the Blackmagic product .

So is 4.4.4 just a luxury not essential?

Your thoughts and advice is much appreciated

.
 
This seems to stir debate every once and a while, I've posted this a couple times, so I'm sorry if I sound like a broken record posting it again. Here is a quote from John Galt, lead designer of the genesis camera, one of the only bayer pattern cameras in the world outputting true 4:4:4. If anyone is biased towards 4:4:4, it would be him. However, In a CML conference, he said - "Based on extensive image composite testing done by myself and my engineering group...It is extremely difficult to demonstrate a improvement even in color difference blue screen/green screen photography (over 4:2:2). The subject matter includes liquids, flame, smoke and graduated color on a fine mesh cloth, all designed to exercise the process to its limits..."

Another fun fact - Ultimatte, one of the leading makers of real-time keying hardware, stopped including 4:4:4 dual link inputs on their products a couple years ago, finding a negligible performance increase over 4:2:2 subsampling.

The fact is, organic source material captured by a digital imaging sensor rarely, if ever, produces hard enough per-pixel color shift to break the interpolation of 4:2:2. 4:2:2 is extremely efficient and accurate at recreating the chroma values that have been subsampled.


10-bit quantization and S-log on the other hand is definitely a huge benefit to have, increasing the cameras sensitivity, dynamic range, and information captured. You can record s-log to a 10-bit 4:2:2 recorder without a problem, and have beautiful footage with nil subsampling/compression problems (given a proper acquisition codec). However I'm pretty sure the sony upgrade will not let you pick and choose, you ether pay for the whole package, 4:4:4 and s-log, or nothing at all.
 
This is coming as a big surprise to me. When I first got my nanoFlash, I spent considerable amount of time testing it - and comparing the 4:2:0 images recorded on my EX1's with the same scenes captured at 4:2:2 by the nanoFlash.

Below is the comparison; as can clearly be seen, 4:2:2 removes the horizontal but not the vertical, jaggies from the red area of the flower. I'd certainly think that in the case of 4:4:4, also vertical jaggies would be removed... Isn't it so? And if not - please direct us to some evidence of the kind I used here...

PS. Please watch the samples atached full size!

flower 420.jpgflower 422.jpg
 
OK, I have posted wrong grabs, sorry - here are the full HD ones; I hope you cannot confuse lower chroma rezultion difference with sharpening, or anything else...

PS. Sorry again - this forum seems to be unable to show full resolution. Please look here:

http://www.dvinfo.net/forum/converg...-vs-4-2-0-example-4-4-4-would-even-nicer.html

These grabs are crops, blown up to 1920x1080. This doean't seem to work here - what am I doing wrong?
 

Attachments

  • RED at 50 Mbps 422 - still H jaggies.jpg
    RED at 50 Mbps 422 - still H jaggies.jpg
    20 KB · Views: 0
  • RED at 35 Mbps 420 - both H V jaggies.jpg
    RED at 35 Mbps 420 - both H V jaggies.jpg
    19.8 KB · Views: 0
Last edited:
moldcad said:
OK, I have posted wrong grabs, sorry - here are the full HD ones; I hope you cannot confuse lower chroma rezultion difference with sharpening, or anything else...

PS. Sorry again - this forum seems to be unable to show full resolution. Please look here:

http://www.dvinfo.net/forum/converg...-vs-4-2-0-example-4-4-4-would-even-nicer.html

These grabs are crops, blown up to 1920x1080. This doean't seem to work here - what am I doing wrong?
Click to expand...

I looked at the bigger grabs, but I still wonder if the differences are just compression artifacts. While they are equivelent data rates (35 and 50 Mbps) because of the different amounts of data they are still different compression scheme (despite both being LGOP) again because of the different amounts of data. As I recall its something like this:

4.2.0
DxDx
xxxx

4.2.2
DxDx
xDxD

4.2.2 still allows you to be a pixel off in almost any direction, but the human eye isn't attuned to slight color differences. I think you'd have to compare it to the same scene at 200Mbps to see if the nearly uncompressed image still had the artifacts.
 
Blackmagic

Blackmagic

henryolonga said:
To get 4:4:4 you don't need to pay that much

If you want to do VFX studio work especially green screen get these affordable cards -

http://www.blackmagic-design.com/products/decklink4k/

or this

http://www.blackmagic-design.com/products/decklinkhdextreme/

This thing is a perfect match for the F3 as long as you don't need to be mobile. Want to go mobile? then yes the Gemini is it.
Click to expand...


Will this work if the camera is on a crane? Will it work with a Very long cable connected to a computer or am I asking a stupid question?
 
you can run HD-SDI over good ol 75 ohm coax cable (typically RG-59) for a good 300 feet before you have to worry about signal locking problems. so yeah, you can tether to a pc a good few hundred feet away with the camera on a crane, dolly, what have you. Did this for a shoot a couple months ago actually, with a single link for an ex1/r
 
moldcad said:
OK, I have posted wrong grabs, sorry - here are the full HD ones; I hope you cannot confuse lower chroma rezultion difference with sharpening, or anything else...

PS. Sorry again - this forum seems to be unable to show full resolution. Please look here:

http://www.dvinfo.net/forum/converg...-vs-4-2-0-example-4-4-4-would-even-nicer.html

These grabs are crops, blown up to 1920x1080. This doean't seem to work here - what am I doing wrong?
Click to expand...

I think what you're also seeing here is an 8bit problem... the NanoFlash records at 8bit... if this was a 10bit file, I believe you wouldnt have the ziggs at all.
 
Hmm interesting document there, I've actually never heard of 4:4:0 before until reading that... with that said, wouldn't a 10bit 4:2:2 image also yield more color information in the red channel making it less prone to jaggies in the first place? I still think it would make a difference regardless if the chroma sampling is different. You still need the color information for it to be effected. Its not like its a jaggie on the image with different colors, its only the red channel which still makes me believe that a 256 vs. 1024 color range would make a huge difference in the overall quality of the image which could be less prone to seeing that pattern. Anyone else have some insight to this? would love to see other information... Either way I still believe that a 10bit 4:2:2 image for effects work, is a very good rate. I have worked on a few films with an FX team attached and it seems that most work in tv is actually being done in a 10bit 4:2:2 color space when getting to the sfx room. I hear all the time about the visual differences between 10bit 444 and 10bit 422, and it all seems minimal to the naked eye. Keying I've seen is very powerful at 422. I think 444 would help with huge amounts of work, something like in a movie "300" where just about every frame is a visual effect, but then again, the way software is now, it keeps getting better and better at handling the footage, with less room for error.
 
Giuseppe,

I absolutely agree that 10 bit would have made a difference - but only "color-wise", not "shape-wise" - i.e. the jaggies would still be there (in anything other than 444), only perhaps less pronounced thanks to being partially filled by additional shades of red - which are simply missing in my 8-bit, red flower example ...
 
Last edited:
I also agree with that, the shape would be there, but the amounts of shades would make it less visible. I think for any visual FX work, the minimum bit depth of 10 for heavy work is standard. You can very much get away with doing FX work in 8bit 4:2:2, but it might take you longer and more tweaking to get the same results as a 10bit 4:2:2.
 
You are making comparisons between 4:2:0 and 4:2:2, in which there is actually a difference that is quite noticeable, however the difference between 4:2:2 and 4:4:4 is very slim. Most importantly, 4:4:4 assumes you want to store color data for every pixel, but how many camera systems are there that can actually capture a signal pristine enough once you start factoring in artifacts like noise and grain? 4:2:0 samples pixels in groups of four, and I can imagine that there is detail fine enough that it could be lost to that kind of compression, but which camera can capture the difference in detail between two pixels and one pixel?
To me, 4:4:4 makes sense if one were adding in rendered special effects or compositing in 3D elements, if at least for the full-res alpha.
 
Kolor-Pikker said:
You are making comparisons between 4:2:0 and 4:2:2, in which there is actually a difference that is quite noticeable, however the difference between 4:2:2 and 4:4:4 is very slim. Most importantly, 4:4:4 assumes you want to store color data for every pixel, but how many camera systems are there that can actually capture a signal pristine enough once you start factoring in artifacts like noise and grain? 4:2:0 samples pixels in groups of four, and I can imagine that there is detail fine enough that it could be lost to that kind of compression, but which camera can capture the difference in detail between two pixels and one pixel?
To me, 4:4:4 makes sense if one were adding in rendered special effects or compositing in 3D elements, if at least for the full-res alpha.
Click to expand...


I dont think anyone here was talking about 420 vs 422... We are talking about 8bit 422 vs 10bit 422 and then 8bit422 vs 10bit 444 those are huge difference in the color information.
 
Giuseppe Pugliese said:
I dont think anyone here was talking about 420 vs 422... We are talking about 8bit 422 vs 10bit 422 and then 8bit422 vs 10bit 444 those are huge difference in the color information.
Click to expand...
I was referring to the sample pictures moldcad posted. He's right in that a higher bit depth wouldn't improve the "shape" of the chroma sampling and going from 4:2:0 -> 4:2:2 will gives visibly cleaner color, but going from 4:2:2 -> 4:4:4 won't be as apparent.
Even in still photography, where you can shoot 14-bit 4:4:4, you still get artifacts which can even vary with the software used to convert the image from RAW and more so once you start applying all sorts of post-processing. So even with a flawless recording format, the weakest link is still the sensor and camera electronics, it's just my opinion but at such high bit depth and chroma all you will be doing is recording more detailed noise on most cameras.
 
Regardless of the math, here is something I can tell you from practical experience.

The more data your fx artist has to work with, the easier it will be and the better the final composite will look.

I do notice a difference between 4:2:2 and 4:4:4 10bit s-log files. Especially for keying greenscreens.

If you can't afford the storage to shoot 4:4:4 slog for the entire show, then at least try to splurge for the greenscreens.

As I type this I am looking at a 4:2:2 log greenscreen in Nuke that came from an Arri Alexa.
Last month I had some Alexa greenscreens of comparable quality that were 4:4:4: log.
Without a doubt the 444 files felt fatter and keyed better.

In this case bigger really is better.
 
You must log in or register to reply here.
Back
Top