I figured I'd post here b/c you guys really take the image apart and probably have best view of subtle color differences.

I'm contemplating capturing out of the HD-SDI port of either a Canon XH-G1 or a Sony EX1. Besides workflow differences and sensor size, a siginificant difference between the cameras is color depth: Canon is 8-bit, Sony is 10-bit.

I would think this should be significant, but I've seen footage comparisons and for the life of me really can't tell where 8-bit was an impediment. The difference just isn't obvious, even though using 256 levels of color instead of 1,024 should be pretty significant.

But I am viewing on a computer monitor w/ untrained eyes, so I leave open the very real possibility that 10 bits is quite an improvement.

FWIW, I'm doing a documentary with mostly sit-down interviews. I hope for a theatrical release, but nothing is certain.

Thanks for any insight!

When you start pushing the colours in colour correction or processing the difference can become more apparent.

8 bit can also cause "stepping" in what should be smooth colour gradients. Things like blue skies, clouds, smoke etc.

For most shots, 8 bits copes very well.

I think the implications of 8 bit material implies a 4:2:0 color space whereas 10 bit implies a 4:2:2 space. There's a lot more info color info available with 4:2:2, which makes color timing 4:2:2 less full of artifacts.

That isn't always true. There are plenty of 4:2:2 8 bit codecs such as DVCPRO 50 & HD.

The bit depth and chroma resolution are independent.

It is a good point to consider though, is the Canon true 4:2:2 out of the HDSDI or 4:2:0 up converted to 4:2:2?

Peter, I've also done my own comparisons between 8-bit and 10-bit. The difference is there, but I didn't find it to be significant for most uses. But, please don't tell my colorist, as he's likely to throw a fit. Whenever I mention "some of the footage was shot on HDV", he gives me a mournful look, like we just had a big disaster. But I do appreciate the additional things that he can do with real 4:2:2 vs. 4:2:0. On a side note, I'm really excited about Scarlet RAW as it lets you choose the best colorspace and bit depth for the material.

Hi

speaking of 420 n 422, is it a good decision to convert your 420 8bit to 422 10bit (something like cineform) video prior to Color timing, so tht colors remian intact n footage has less artifacts? as i know there is no advantage, anyone has anything to say?

Thanks

The matter is more complex than simply stating 10-bit color is better than 8-bit. Your question is ultimately about quality. And overall quality is composed of different components, visual as well as technical.

If you focus on the technical side you can say that any "higher" spec is better, like better subsampling (4:4:4 > 4:2:2 > 4:2:0), better color quantization (8 vs 10-bit), higher bitrates or better - newer, faster and more efficient - codecs.

Quite often you will find yourself in a 'format' discussion about all combinations of these and the most important parts are left out: the lens, the sensor and the (in-camera) processing pipeline.

Older camera's often have less sophisticated sensors that record less data in the first place. That is information you never get back, whatever you do.

Then there is the processing pipeline; high(er)-end camera's may use 14-bit A/D converters for precision, other may use 10 or 12-bit (it should always be higher than the sensor quantization and the output).

What is "eventually" output for recording may be 8, 10 or 12-bit. And only then does the recorder/encoder come into play.

And even there things are not so clear cut. 50Mb/s of anything is not always the same quality. Newer implementations of (hardware) encoders - using the same codec - are simply able to do a better job than previous generations. New codecs may also be better than older codecs, but possibly not for all applications.

And lets not forget about the lens! This is the first component in the system that affects the image quality. It relates closely to the sensor. In short: it is much more difficult to make a lens with a high resolving power for small sensors. So a smaller sensor generally means a worse lens.

For comparisons you really have to know what you are looking for and then compare 'systems' as to their individual performance. Manufacturers don't publish enough information to make any useful assessment from the technical specs.

The sensor is IMHO the biggest factor. Bigger size and full resolution will give you more information and if it is coupled with a good A/D converter and records using a modern encoder/codec, with a reasonable bitrate, you should have great visual quality results.

Higher bitrates will yield increasingly less visual improvements. 4:2:2 will give you more information but is only really useful if you need serious color correction; the same pretty much goes for 10-bit quantization.

If you (think you) need 4:2:2 over 4:2:0 or 8-bit over 10-bit you really need both. For a feature film shot on different sets/locations and times, 4:2:2 and 10-bit quantization may be requirements to make it easier to bring everything together. For interview and doco footage this may be less so.

For color correcting 8-bit footage you can use a 10-bit codec in post to preserve as much information as possible.

George/

Thanks VERY MUCH for the replies!

BTW, as for 4:2:0 versus 4:2:2, with either camera I'd be capturing out of the HD-SDI port, so would by bypass their respective codecs. And single link HD-SDI is 4:2:2.

If you plan to drag uncompressed footage from the EX1 you best have a super fast computer and lots and lots and lots of drive space.....

The best way to get that footage from the EX1 is use a XDR recorder... and even at that the footage is still compressed...

Ray,

Thanks for your input. The plan is to use a fast workstation w/ a large RAID array for capture. For the sit-down interview portion, mobility is not an issue.

If this proves unwieldy, then I'd probably use a 10-bit color codec like Cineform. In which case, image degredation due to the compression/decompression will be pretty minimal.

don't forget that most LCD screen are 8 bits, so disply will utimately been done on 8 bits.

Right, which really makes me think that the difference is more significant than I'm seeing.

You might get some ideas over at the Black Magic web site... they have products for capturing uncompressed 10 bit data...

yes, I like Cineform very much... but I think you will need some hardware to go along
with Cineform, to work the way you want..

don't forget that most LCD screen are 8 bits, so display will utimately been done on 8 bits.

True, but if you consider printing to film it is worth it. You'd better use a display like the HP DreamColor LP2480zx color-critical (10-bit) display. On the road you can take the released HP EliteBook 8730w with you, which has an integrated 17" DreamColor display (only 8-bit tough).

Mobile workstations for the ultimate creative professional

Designed for the most sophisticated graphic-intensive applications, the new HP EliteBook 8730w is the only mobile workstation on the market to offer a 17-inch diagonal HP DreamColor Display, which offers over 16 million colors, a significant improvement to the 260,000 available colors on traditional notebook PCs. The notebook also offers an Intel® quad-core processor, 8 gigabytes (GB) of memory (64-bit OS needed) and NVIDIA’s next generation of Quadro FX cards with up to 1 GB dedicated video memory.

George/

i think HP is making a feature out from nothing with their laptop.
yes, while most PC screens are TN LCD (with 6 bit color) there is a huge quantity of screen
made from VA technology (MVA, PVA, S-PVA) or IPS (and S-IPS) with 8 bit color.
So assuming THEY provide 8 bit while the rest of world is stuck to 6 bit is a bit strong.
Last year , Apple was almost sued by doing the same kind of advertising with their new laptop.

I´ve made a filming not long ago to transfer to cine. We used an Canon XLH1 and captured with a BlackMagic directly to the computer in 4:2:2 10 bits uncompressed. we monitored at the same time through a blackmagic HDLink to a cheap Samsung 24 inches 1920x1200 screen. Man, you could see every single pixel!! For that I had to put four SATA (500 Gb each) in stripped raid (it was the only way to get enough speed for the capture, in fact with two or three disc the speed was ok but when the discs start to get half full the transfer rate drops dramatically). To make the color grading I used a NEC Spectraview 26 inches calibrated with an Greta macbeth eye one. The beauty of this monitor is that, even if it 8 bits, is wide gamut (shows the full adobe rgb gamut) and the calibration is stored in the monitor hardware. I used the cineon LUT as the target (I tried to get the LUT from the lab to match their printer but it was impossible, why? is a long story). I made a very extreme color grading full of secondaries (it was pretty easy to qualify for the secs). The quality was superb. At the lab one of the engineers asked for details thinking that it was made on film. In the other hand the 35mm printed version matched the colors and contrast incredibly well it was really WISIWIG.

Martin...

I'm a little surprised at your use of the Gretag Macbeth. The color mapping this tool generates is designed to replicate, as close as possible on a computer monitor, paper printer colors, not film colors. If you used a 3rd party LUT, that may have overridden the color map from the Gretag, and kept you from further problems.

Bill,
I used the Eye one together with the spectraview profiler software. The instrument (in this case the eye one) is only the interface for the software to read the colors from the screen. The soft "talk" with the screen hardware and make a software/hardware calibration. You can choose whatever the LUT you want as the target and the system try to match that LUT as much as possible and then dump that LUT to the monitor hardware plus an extra profile on the operative system if you like. Of course there are higher end instruments but the eye one worked great in this case.
I made some tests and bring to the lab (Egmont Nordisk Lab) in Denmark. We compared the tests between my system and their flame station with two different monitors (calibrated by Arri tech service), at the monitor in the printer room and at their Color Grading room with a CRT broadcasting monitor withy their LUT and then the printed version in 35 mm and the material always matched perfectly (it was hard to see any difference)

I forgot to tell you that i didnt use any woking lut, i just used the calibrated monitor targeted to the cineon curve. For more details I worked in linear. I made some tests going to log but the difference was not enough to make it worth for that specific footage.

I´ve made a filming not long ago to transfer to cine. We used an Canon XLH1 and captured with a BlackMagic directly to the computer in 4:2:2 10 bits uncompressed. ...Martin, and yet from everything I've read the XL-H1 does not spit out true 10-bit color. It's 8-bit color. The last two bits are padded with 0's to meet the HD-SDI spec, but they're meaningless.

That said, I've seen amazing color from the XL-H1, which makes me really wonder how much those extra two bits provide.

I really cant tell you how much that two bits provide. What I can tell you for sure is that the footage coming from the tape and the same footage coming from the SDI when color graded with exactly the same settings (remember that it was some extreme C. grading) the picture coming from the tape showed some banding and the one coming from he SDI was absolutely free of banding. May be was the difference on chroma sampling? (4:2:0 vs 4:2:2) may be the 10 bits vs 8 bits? i dont know.

I'm also sure the XL-H1 is 8-bit too.

However, that doesn't detract from the footage since i'm sure the footage Martin was seeing is incredible! 4:2:2 8-bit uncompressed is beautiful - though it'd be even more wonderful to work in 10bits ;) I can attest, I *never* see banding with the uncompressed 8-bit footage I've shot (I have a 14-bit out to my analogue HDTV and I can't detect any banding on it either).

Martin, is there anyway I can see some clips/stills/footage from this project you worked on?

I've been shooting uncompressed for a year now. I just did two videos in the last two weeks that were shot uncompressed HD-SDI to my Blackmagic card - shot from a Canon HV20 and the results are impressive - would love to see others' results as well!

I'd like to see someone's results of shooting uncompressed with the 10bit EX1.

Christoffer,
I have an Ex1, tomorrow is returning from the service (update of firmware and so on).I will make a test capturing 10 bits uncompressed and post the images. About posting the images from the film with the XlH1 i will ask the director for permission since the movie is not released yet.

Martin, that's fantastic - thank you! Now I'm all excited to see 10-bit .PNGs!

i´m trying to post full size pngs but the system doesnt accept it. I´ll try another way.

Thanks Martin!

ok, here we go: Untitled Document (http://evengod.eu/shoot/pics.html)
this pics are from the movie Storstad (the big city). The footage was captured uncompressed 10 bits with a Blackmagic from the SDI of an XLH1 fitted with a Mini35 . The lenses were from a Leica prime set. For the lighting i used only two Falcon eyes (cheap Kinoflo) 200W each, 2 softboxes 1000W each, 2 750W fresnels and one 1K fresnel and some pieces of 1/2B filter plus some pieces of diffusion paper.
The Color grading was made in Apple Color and the compositing in AE (for example the fan I added in post, is an animation made from stills of the fan).
In some pictures you can see some banding, that comes from the grabbing of this pics, the footage doesnt show any banding.
I made also some tests with the Ex1 to compare the material coming from the SxS and the Blackmagic capture but i´ll post later on because I cant upload it here.

here SxS vs SDI (comparison) - The Digital Video Information Network (http://www.dvinfo.net/conf/showthread.php?p=931005#post931005) is the first pic from the EX1