There seems to be a lot of theoretical talk and speculation in regards to 4k 8-bit to HD 10-bit subsampling. My question is, do any of you have actual proof that it works? Have you personally tested it? It should be quite simple to shoot a gradient, push the grade until it breaks and publish the results.
FS5: FS5 4K color and bit depth
- Thread starter naavt
- Start date
Did you read the linked article?
I've just tried it before reading your post. I've downloaded some GH4 4K footage that Philip Bloom posted on Vimeo, converted it to ProRes 444 and tried to convert it to Cineform. Don't know why but Resolve keep giving an error on the last one. So I've opened up a 1080p timeline and have made a key to the sky and the key seems exactly the same on the 4K and 1080p footage.
Probably I'm doing something wrong but I have to say that it didn't work for me.
If I had to bet I would say:
C - Real 10-bit HD image
B - This one's sharper than A so I say 8-bit UHD to 10-bit HD
A - Original 8-bit UHD
Prores is YCrCb. You have to convert to 10 bit RGB format minimum, or Cineform 12 bit 4:4:4 RGB. The linked article quotes David Newman how it can be tested. No one including David himself has said it would be visibly improved. It is academic.
JMtheDP
Veteran
I'm sure the UHD-HD proponents will have no difficulty in picking out the real 10 bit HD image, the 8 bit UHD to '10 bit' HD image and the original 8 bit UHD.
Hardly a fair, or meaningful test as a large part of the 10-bit issue is malleability in grading. Also, none of these three pictures line-up, at all.
Nonetheless the image C presents a quite obvious smooth tone transition in comparison with the other two.
Nirv
Carbonite Member
LOL, do please tell how this isn't a fair test and what my process was which you appear to be able glean from the image I posted. I'm all ears.
I went out of my way to make this a very fair test for the record.
Sorry you don't like the results, let me know your bank details and I'll give you your money back.
Yep.. As I understand it, the science is sound but I see no real world application of the technique demonstrating a clear advantage of sub-sampling(or super sub-sampling) higher resolution to increase color bit-depth.
Barry Green posted several grabs on a similarly themed thread on the GH4 forum.
JMtheDP
Veteran
I simply mean that the A/B/C displayed here are not all evenly lined up (which makes me curious about the methodology) and I believe the issue that the OP has is that he's going to have a massive variety of "input" to deal with.
Judging one small portion of one frame of black and white footage is not exactly conclusive testing for a three year project. Not that we don't appreciate what you're doing, but I would be more interested in knowing the exact procedure you're using to achieve your results.
For what it's worth I prefer picture C and I imagine that is 10-bit HD. I imagine B is the original UHD 8-bit picture and A is the downconverted picture.
I used to shoot quite a bit of 8-bit HD when I used HDCAM, but I never used log with it. I used some hypergammas and I found even those could break under heavy grading.
Nirv
Carbonite Member
I apologise unreservedly for my slapdash photoshop skills in throwing together this example late last night but the important aspect of the test was the separation of the 8 bit samples under grading which A and B clearly show.
If a 3 year project is important enough then anyone involved in it should be renting the camera and battle testing it for themselves and not relying on a hack on DVXUser. I do my own testing, I did these tests months ago and I would expect others to do the same and would urge them to. Naavt seems like an experienced capable person who could get all the necessary data from a visit to a local dealer with an SD card if he or she wished.
As I said in my initial post, the maths are sound but they don't survive contact with reality and there has been a lot of hyperbole written about the technique on some blogs and forums like it's some free lunch which it clearly isn't. The reality is that all the factors which limit grading remain in the down sampled image be they compression artefacts or slabs of 8 bit posterisation.
Grug
Veteran
I'm confused too now Tom, because in your reply to me you refute that you can't get 10-bit from downconverted 8-bit 4k:
But in other posts clearly restate David's points that you can't actually get 10-bit from downconverted 8-bit 4k, merely 10-bit precision in one of the three RGB channels.
So I'm assuming that, actually, we agree... but if that's the case, I'm confused by the change of viewpoint in your earlier response to me.
Unfortunately that was also my initial founding after making the first experiment. Tom Roper commented that I have to convert to an RGB format instead of YCrCb ProRes though. I haven't tried it yet but I will. Regarding the camera itself, although I would prefer FS5 smaller footprint and weight, I do think that FS7's main advantages far surpasses the size and weight it looses for it's baby brother comparing the 900€ difference that separates both cameras.
I also must address some attention to both camera rolling shutters abilities to deal with moving stuff since most of the project footage will be very action driven.
starcentral
Veteran
You really have to read between all the lines in this 4k 420 -> HD 444 stuff, and I can see why it's so easy to get into disagreements over it.
Since 420 means you have lower resolution sampling of color then for sure 100% when you downscale 4K to HD you can end up with 420 -> 444. I'll buy that for a dollar.
You can also take an 8-bit image into 10-bit or higher bit-depth and reap some benefits when manipulating the image, but it will absolutely never ever be the same as if you had shot it in 10-bit (ie. 1024) levels to begin with. The original gradations are just not there.
To me it's like the folks recording 8-bit signals from their Sony A7S with 10-bit recorders and telling me it's the same as my F55 is recording. Er… um.. . :shocked:
Since 420 means you have lower resolution sampling of color then for sure 100% when you downscale 4K to HD you can end up with 420 -> 444. I'll buy that for a dollar.
You can also take an 8-bit image into 10-bit or higher bit-depth and reap some benefits when manipulating the image, but it will absolutely never ever be the same as if you had shot it in 10-bit (ie. 1024) levels to begin with. The original gradations are just not there.
To me it's like the folks recording 8-bit signals from their Sony A7S with 10-bit recorders and telling me it's the same as my F55 is recording. Er… um.. . :shocked:
abreu-canedo
Well-known member
The 420 to 422 part of this equation seems sound. I totally get it and agree with it. However, the turning 8bit to 10bit seems suspect to me since you would not add more colors by dow converting. You simply get to throw away some of the slave pixels in 420, thus turning it I to 422. But to say you can go from capturing 255 to 1000 different colors by taking 4k footage and bringing it to 2k? I'm missing the logic here. 420 to 422 I'm on board. But perhaps I lack the knowledge to understand how you would go from 8bit to 10bit. After all, those are separate things and you can have 8bit 422 or 8bit 420.
Nirv
Carbonite Member
I would suggest you think about the lightweight FS5 with a decent bit of heavy glass hanging off the front and consider whether that is in any way an advantage? If you're only going to put Sony's flimsy E Mount lenses on an FS5 it'll be fine but put a 24-70 F2.8 on it and it'll be so front heavy hand held shooting will be uncomfortable to say the least.
I think you'll more than get €900 extra value out of the additional features of the FS7 and you can shoulder mount it without any rigging if you rig the VF much further forward which you can't do with the FS5 which is aimed squarely at hand held shooting out of the box.
The FS7 rolling shutter is certainly adequate but not the fastest around, I rather doubt it'll be any issue for you unless you shoot like a whirling dervish.
JMtheDP
Veteran
There are good deals on used FS7s beginning to appear too
http://www.reduser.net/forum/showth...7-package-for-almost-loosing-2500-(crazy-deal!!!!)
http://www.reduser.net/forum/showth...7-package-for-almost-loosing-2500-(crazy-deal!!!!)