F3: Trouble with 10-bit

my vision is not perfect, and i am not a pro colorist...... but i look at 2 images posted above and i do see the difference between them. look at the hairs on a face.... one image surely looks much sharper then the other one, a lot more small details are visible. the only thing i know for sure about green screen - lighting is the most important thing, its the only thing that separates the object from the background! I have seen just PERFECT green screen shots done with panasonic af100 on internal avchd codec, where female was playing with her hairs.... so i know for sure lighting is the secret key for green screen.
 
Last edited:
Can you post tiffs please? Jpegs are 420.

I've keyed 422 and 420 and can say without a doubt that there is a difference. It's not THAT big but there is a difference. And as somebody else pointed in this thread your subject matter isn't that complex.

Also can you post two original frames.
 
OK, here's another test in the form of pixel-peeping. Now I can see a difference, particularly with the pink lower tooth! And the generally increased level/faithfulness of detail. But am I seeing the effects of the 4:2:2, or the 10-bit? Both, I guess. I suppose that the pink tooth and pinkish eyes, for example, are a result of 4:2:0, whereas the grade of shadow across his thumb is a result of bit-depth.

Ironically, I think that the extra fuzziness introduced by the in-camera codec actually IMPROVES the quality of this particular key. But I would be willing to concede that may be because of my skill level, from lighting through post.

http://imgur.com/a/OKOQr#0

These images are all full-raster .png screen-grabs at 400% from Premiere.

In this album, the first image in each pair is 8-bit, 4:2:0 XDCAM-EX recorded internally in the PMW-F3
The second in each pair is the 10-bit, 4:2:2 ProRes recorded via SDI on the Blackmagic Shuttle.

What kind of errors am I introducing by performing screen-grabs in this way? KINOKS educated me just now on one difference between .jpg and .tif compression. At any rate, whatever problems a .png screenshot may create, there is still a visible difference at this magnification which makes me feel a little better about using the Shuttle.

EDIT: By turning my attention to "blocking," the difference in skin rendition becomes so glaringly obvious. In the close-up of his eyes, practically his whole face is covered with squares! But is that the 8-bit at work, or the 4:2:0 compression?
 
Last edited:
Finn Yarbrough said:
EDIT: By turning my attention to "blocking," the difference in skin rendition becomes so glaringly obvious. In the close-up of his eyes, practically his whole face is covered with squares! But is that the 8-bit at work, or the 4:2:0 compression?
Click to expand...

Neither:) That's the 35Mbs mpeg2 compression.
 
Finn, with all due respect, all the attention your question has raised is not worth this much energy because its a bad test. Your casting doubt on all the standard wisdom of the value of 422 over 420 with a shot that won't reveal it. You haven't even told us what camera you've used which could be very pertinent.
Unfortunately this kind of discussion is all too common on the web, raises lots of commotion and reveals little.

Look i could post a tight closeup in super 8 and it might look fine compared to the same shot in 65mm, but if you used the same cameras and shot a landscape the difference would knock you over the head. Actually on a small web page you might not even notice the difference even on the landscape , but if you blew it up your jaw would drop.

There are people who will swear they get great green screen with a 5d shooting heavily compressed footage in 420 - and I'm sure it is fine in some situations - but not in more difficult ones. There's no question you can get away with shooting some green screen in 420 . The question you should be asking is when will it make a difference, when do you need the external recorder.
I wish I could point you to the tests I saw on the web somewhere . In those, which stressed the codec, the difference was tremendous.
 
Last edited:
Lenilenapi said:
your question has raised is not worth this much energy because its a bad test.
Click to expand...

The bigger question is worth considering IMO as a recorder is such a PITA and the extra data load is too..

I think there are many things going on - maybe bigger than the 'space' of the images, mainly the compression level.

Sony are genius (in an odd manner) and I guess their compression offers great image per gb.

Also worth considering is sharpening - do you need it and does it break the image.

Overall both images seem fairly crappy to me - I think green screen is somewhere where 4k acquisition or even tipping the camera on its side where possible is a big boon.

Id still wager that the 10bit, colour corrected, sharpened and in a gradient intense scene would win..

S
 
Lenilenapi said:
Finn, with all due respect, all the attention your question has raised is not worth this much energy because its a bad test.
Click to expand...

This is true. But the point of this test as I see it was not to stress the codec. But to see in some "normal" circumstances what the difference is between two compression types.

Anyway...:
In general this is how it goes:

- Bitrate: will effect grading the most. A low bitrate codec will far apart very quickly. This will happen way before you start noticing the effects of 8 bit banding or 420 jaggy edges.

- Bitdepth: will effect banding. This can be seen on flat surfaces like a blue sky or a white wall if it has a gradient lighting. It can also be seen if using a very flat gamma when you expand the range in post. This would show it self for instance as plasticity skin tones or as general image quality degradation... Won't affect grading in general if post is presented with a relatively large bitdepth codec.

- Chroma subsampling: will effect contrasty edges the most. For instance a red T-shirt over a green background. Will effect other thin lines, especially those with a strong chroma component. Won't affect grading in general if post is presented with a relatively large bitdepth codec.

- Temporal compression: GOP will fall apart when pushed in post if the material doesn't have a large enough bitrate. When shooting with the F3 in Slog to the XDcam codec I had trouble relatively quickly with macrobloking when I pushed the codec. But it's when I added saturation that's when the image fell apart the quickest. Just pushing the levels didn't break it as quickly as adding saturation. Try it yourself! LONG GOP will also effect moving images with lots of details IF bitrate won't be high enough!

So as you can see small bitrate is generally the culprit but chroma keying is a special scenario. All the bitrate in the world won't help you with jaggy edges if you are shooting in 420! Now, your example isn't really stressing the codec, as has already been said in this thread. The biggest issue being that your subjects Tshirt doesn't have a strong chroma component. He is dressed in black!:) Do another test and dress the talent in red or orange, etc. and you'll see a difference. If you Can, try shooting in 444 and it will be that much more obvious. Then try putting a blonde in the footage that has lots of tiny hair and that has a semi transparent blouse and that moves a lot and then have fun!:)

So, in some scenarios the difference between 420 and 422 or 444 for that matter will be much more apparent than in others. Your scenario is one of those that a 420 compression is more or less enough.
 
KINOKS said:
Then try putting a blonde in the footage that has lots of tiny hair and that has a semi transparent blouse and that moves a lot and then have fun!:)
Click to expand...

Or just film a blonde in a semi transparent blouse and don't even worry about this whole sub-sampling malarky...
 
Lenilenapi said:
Finn, with all due respect, all the attention your question has raised is not worth this much energy because its a bad test. Your casting doubt on all the standard wisdom of the value of 422 over 420 with a shot that won't reveal it. You haven't even told us what camera you've used which could be very pertinent.
Unfortunately this kind of discussion is all too common on the web, raises lots of commotion and reveals little.
Click to expand...

Lenilenapi, this entire discussion is posted under "Sony F3/F5/F55," and the title of the thread is "F3: Trouble with 10-bit."
I have posted both a wide-shot comparison as well as the close-up pixellations. The shot is of a green-screen chroma key, which is exactly what standard wisdom claims necessitates the use of 4:2:2.
It's OK if you don't have the answer, but you should maybe read the thread more carefully before trying to shut it down.

On another note, KINOKS, your comments have been very educational to me. I think that the conclusion I can draw from this (apparently mostly useless) real-world test is that, for this particular job, I can save the space and effort and keep working in-camera. When we start our next documentary, though, where there will be a lot of pre-dawn, sky, and landscapes involved, the Shuttle will be ready.

Thanks, everybody, for your 2-cents.
 
Finn,
Sorry about missing the F3 note at the start, and I don't mean to criticize you for posting in the first place, or to get personal. I just found it frustrating. If your question was limited to: can I get away with shooting to the 420 cards on a simple green screen without frizzy hair - then by all means many people would say sure and indeed your test implies that.

But unless I misread, it seems you were questioning the whole idea of whether 422 really does make a difference in green screen - all based on your test. That's what most of the discussion has been about and in that case your test just isn't very meaningful.

I think you have drawn the right conclusions for your specific job though so I'm glad the discussion helped.
Kinoks that was a great answer very useful info.

BTW Is anyone else familiar with the test i've seen on the web that includes people blowing cigarette smoke ( that shows off 422 vs 420 for sure). I can't find it.
 
Last edited:
You must log in or register to reply here.
Back
Top