View Full Version : Si2k and greenscreen
December 4th, 2010, 09:42 AM
Alex, Rohan, any other SI2K operators, any clues for greenscreening with the SI2K would be appreciated rather urgently. We have a shoot on tomorrow, only managed to get the screen rigged and de-wrinkled and lit today.
We have a screen adequately and evenly lit, a Canon 7D captures a briliant green, the SI2K's green is quite dull, almost undersaturated. We can get a "fair" key but hair separation eludes us.
For a dumb idiot like me, where on the false-colour exposure display would you set the green? We have it in the gray presently. Should we bring it up into the yellow?
Our first tests with daylight corrected lighting we so-so, not as clean a separation and a bit of noise in the background plate which when it is taken out then appears in the main image.
We put green gel on the lights and that improved things but there is still something not quite right. The green is still quite undersaturated.
Any assistance greatly appreciated.
December 4th, 2010, 02:08 PM
Have done a bit of greenscreening without issue (except the first one which was shot the day after I first received the Mini). In fact one of the production houses said it was the easiest keys they ever pulled.
What program are checking it on?
It sounds like you are doing the right things with an evenly lit screen - I would bump the exposure up, it does sound like you are a bit under if you are getting noise.
Keep doing those black balances.
Make sure your subject is separated from the background so there is no spill, especially if you have green gels on the lights. Of no help to you right now but I have been lighting the background with KinoFlo green tubes - they are brilliant.
What Look are you using? I've done most of mine with the default - yes the colours don't 'pop' like the over saturated DSLR but it's clean.
The in camera keyer is great for checking composition - it's not a 'perfect' keyer but it gives you a really good indication of whether your foreground is separated from your background or not.
If your background & subject are well lit there should be no problem pulling a great key.
Give me a call if you need to (although without seeing the set up it's a bit hard to offer specific advice)
December 4th, 2010, 04:22 PM
Thanks for your prompt reply.
Yes. I am using the default look and black balance is something I find I should do often. We did do the tests with a camera black-balanced at startup but I did observe in the keyiing a little of the horizontal banding in the noise, so we had probaly gone about 20 minutes into the test without re-doing it.
December 4th, 2010, 06:50 PM
It is imprtant to not do a black balance immediately on camera head power up. The camera should be powered on for a few minutes to allow it to reach a steady state temperature. If the temp is not varying, such as indoor conditions, there shold be no need to rebalance again.
December 4th, 2010, 11:46 PM
Bob, sorry - didn't see your message in real time.
SI2K produces one of the cleanest images among the video cams, so you really should not be getting any noise issues.
What you see is probably more like blotches of the matte from the uneven lighting after keying.
I personally never light with green gels/tubes because the green spill is hard to fight the way it is.
Instead, I use fluos in softboxes or bounced off umbrellas. They usually create a very soft lighting that is ideal for greenscreen.
For the greenscreen itself, I do not like using fabric - it is always too wrinkly. Rather, I'd recommend using paper roll greenscreen - gives me great results.
Check B&H for Savage brand (http://www.bhphotovideo.com/c/product/45636-REG/Savage_46_1253_Widetone_Seamless_Background_Paper.html/BI/2187/KBID/2932). Since you are down under, you probably have your local supplier that sells an equivalent product.
Try not to put your subject any closer than 4ft to the greenscreen, or spill will haunt you in post.
Green screen should be lit 1-2 stops darker than the foreground action/objects.
As for pulling the key. Use After Effects with Keylight and Key Correct.
Do NOT expect to pull clean matte by just using one key on everything - this is usually impossible. Instead, you'll have to treat different areas of the image with different key settings.
Hair is of course hard to process. Anything fuzzy produces a conundrum for the keyer.
However it can be done - look at my site for the Topher's music video Hero. The Japanese girl's long hair is flying in the wind with the office window panel in the background. That was pulled from a greenscreen matte - there was no office, just a greenscreen for everyone - and Topher and the girl were never filmed together - although you can see them composed in the same frame in some shots.
December 5th, 2010, 03:33 AM
Thanks Alex and Rohan for your advice.
We went ahead and shot anyway and brightened up the greenscreen a bit. Here's what we now have. We also used a wonderful Canon HDTV lens which gave us just a bit more than the Angenieux T2.2. Gavan shot with it outside of the range it vignettes.
December 5th, 2010, 11:08 AM
Photos show that greenscreen, unfortunately, was not lit evenly as it should. The 2nd pic's back window will be rather difficult to pull matte on.
I do not see hair as problem in your example, unless it was flailing in the wind. Yes, keying will probably kill a couple of strands but you can make it look natural if it is uniform throughout the shot (not like hair is appearing/disappearing.)
Also. Seems like consensus is that the greenscreen should be lit about 1-2 stops darker than the foreground. I'm not sure that was the case on the set in question.
Also. What's with the very obvious banding on green, and with the severe aliasing around the kid driver's nose in 1st shot etc. ? This doesn't look like SI2K... Did he use Cineform RAW on FS1 or FS2, or on Adaptive setting??
December 5th, 2010, 11:43 PM
Hello again Alex.
Australian rules - kid is passenger, right-hand drives over here.
I got caught out badly with the rear side windows on the Toyota Echo. They have about a half-stop neutral density tint in the glass and I did not see it. This was one problem area in an earlier test.
The banding ? I did not see any on Premiere but it is likely there to be seen on a larger display. The time between set blacks drew out a bit and the cam temp went up to 46 in the hot shed we shot in. This camera ( number 178 ) seems to be a bit more prone to this than the other newer camera over here.
The aliasing. That ambushed me a bit. I think it might be related to using the Canon HDTV B4-mount lens, which fits up to the camera via a special IMS mount. The mount contains a big piece of optical glass which mimics the effect of a prism block. There might be CA there.
The lens also seemed to be soft when opened up, zoomed in to about 132mm and looking through two 4x4 filters, Tiffen ND9 and Schneider Tru-cut.
Camera settings were 2048 x 1152 @ 25P, daylight preset, adaptive setting.
This is my first adventure into chroma-keying.
December 6th, 2010, 12:20 AM
What was your gain setting? I always shoot at -3dB gain to minimize noise.
CFHD RAW compression setting: Adaptive - why? This can give you inferior results. Why not force it to Quality3, which equals to Cineform Filmscan1. If your system cannot handle it, then it is not suited for production. Adaptive is just not good enough for guaranteed quality greenscreen work...
Driver: put him back in the left-hand seat where he should be. This critically affects the matte quality....OK, messing with you here :)
But hey, just relax and use what you got, you'll be fine with multiple keys and garbage mattes.
December 6th, 2010, 08:15 AM
Normally I shoot at -3db gain. On this one because of our limited power budget for lighting, the camera was on 0db gain.
Adaptive? I admit I have not tried the other settings in a good while. The cameras have been used a few times for long takes like a music event. I should test again as I now have a SSD in a dataport.
Driver in the left seat? Fix that soon enough by flipping the image. We really enjoy the benefits of tech these days.
We have the traditional takes, in this case because of a third person in the car, a medium wide, crowded two-shots, close-ups and reverses. The shots with rear side window won't be dominent.
The recorder in use is the P+S camera body.
Thanks for your input. Yes we let the camera warm up for about 15 minutes whilst we set lights. In hot conditions we shoot in here, there can be a variation in the camera temp if the takes are a bit long. To be on the safe side, I black balance if the temp changes up and down more than another ten degrees after warm-up.
The DOP, whose history is in film, distinctly likes the images from the SI2K for their dynamic range.
December 7th, 2010, 10:26 AM
Interesting that you both shoot at -3Db when you can, does this significantly reduce noise or is it to reduce depth of field? I've never really tested it as I was under the impression that it increased the noise slightly ( I am not sure where I heard that...) as it was a movement away from the camera's native ISO of 160. Am I missing something here?
December 7th, 2010, 08:12 PM
What's the time of day and setting for the shot?
December 7th, 2010, 08:39 PM
Interesting that you both shoot at -3Db when you can, does this significantly reduce noise or is it to reduce depth of field? I've never really tested it as I was under the impression that it increased the noise slightly
If one can win "slight" noise reduction by lowering gain, one should do it :)
DOF cannot be affected by that - electronic gain in signal processing chain is not the same as lens aperture.
...or did you mean that we are forcing the lens aperture opening this way? Nah, this is just to get a cleaner (less noise) image.
December 8th, 2010, 12:39 AM
What I meant is that employing -3DB, effectively reduces the sensitivity of the camera, which allows you to reduce the t-stop, so the aperture is wider open, hence less depth of focus (?)
If you gain ( maybe that should be 'achieve') less noise at -3DB is there a payoff anywhere else in the image? - sorry if this is elementary stuff, but my background is in film.
December 8th, 2010, 07:18 AM
I would like to claim to be a wise old man but truthfully, with -3db, I am slavishly following common doctrine relating to more conventional video cameras with a view to minimising any sensor noise.
I have been accustomed to slower stocks or tungsten balanced 320ASA with dalyight filter, so I guess it is me trying to remain close to what I know. I have also been playing with 35mm groundglass adaptors on video, which also perform closer to slower stocks light-wise.
The SI2K does not have any inbuilt ND filters. There are circumstances where one wants to be lighter weight and the mattebox and filter set stays off. In that event -3db is desirable to be as little into the diffraction zone of the lens as possible when having to close the aperture.
This is speculative on my part but -3db may also provide a hedge against sensor noise if one forgets to black balance after a significant camera temp rise.
The setting is morning at about 0830 Perth summer time. The juvenile offenders are headed for their hearings in Children's Court after having been on supervised bail.
To be consistent with the lighting, it is intended to shoot the background plates from a vehicle travelling on northbound streets and freeways in the early morning.
There is also a bit of a stylised mood thing going on with the lighting and colour grade when it happens.
I cannot claim any credit for the lighting scheme. I know only just enought to get myself in strife so have had serious help on this one.
Key was 1.2K HMI, gelled half CTO I think.
Rim lights were 650watt tungsten fresnels.
Greenscreen were 3 x redheads with green gels.
Available power was a bit limited - 240v single phase supply to house only.
Regards the wrinkles in the 20' x 10' greenscreen, DOP Gavan was onto a good trick. He has one of those Danos Direct "wait there's more" little steam cleaner gadgets with its own small vacuum cleaner head on a hose. I had my doubts but they were unfounded. After the fabric screen was hung and pulled to mild tension, we ran the stream cleaner over it and the wrinkles straightened out. I thought the creases would remain however they pretty much came out as well after the fabric dried off.
December 11th, 2010, 03:16 AM
Bob, good luck with this and I'm very curious to find out how the key went.
It looks to me like the car could have been lit more brightly, and a little less warm. Sometimes it's a good idea to be sure the talent's hair is a little more under control, e.g. girl with the dirty blonde hair.
But I think you should definitely be able to get a good key out it. And I think having a moving background may make selling the key easier, although you will have to match color temps and pobably lighten some areas on the talents' face. Just a guess.
Thanks very much for sharing, and if you get a chance, let us know how it all turned out ;).
December 11th, 2010, 10:04 AM
I think it was that I did not have the half CTO gel for the 1.2K HMI.
I believe there is also a style thing happening. I think the backgrounds are to be oversaturated and contrasted. Power resources were also very limited in the shed we were using as a studio.
I'm trusting my DOP. He has about 30 years of film lighting experience in 35mm commercials in AU and offshore in Singapore and Jakarta. He tells me there is a technique called green suppression which he has seen used in similar circumstances for commercials.
On a previous shoot, he created stuff out of the available light I was already set to give up on.
Girl with dirty blonde hair? The young guy might be a little concerned about that observation.
December 11th, 2010, 11:10 PM
If that's not a girl in left of the second still you posted (post #6) then I've become too old for this business, LOL! :)
December 11th, 2010, 11:26 PM
The boy is dirty blonde.
The girl is clearly blonde.
December 12th, 2010, 11:46 PM
No clean or dirty female blonds in there. The two fairheadeds are Cameron and David, both minors playing the parts of crimmy teenagers on their way to juvenile court, so no last names without parental consents.
You have it correctly Peter and they say that making movies is ageless. Lisa Bennet has the dark hair. She is a true professional and very committed, enquires in detail about her role and the backstory behind her character, then "owns" the character entirely.
December 13th, 2010, 05:28 PM
1. Is this V1.1 or V2.0 - big difference in compression quality. Judging from the stills it should be quite easy key in compositing, not one click thing but nothing really complicated in nuke for example. Just make sure You use advanced detail2 debayering, or do sharpening in Your compositing software.
2. -3dB does NOT make less grain, it just limits the highlights never reaching 100% so it actually just clips dynamic range - not good for anything that I know....
3. If you are shooting things slightly underexposed (lots of dark areas in the image) it is actually better to shoot +3 to +6 to get clearer image, but should not necessity in this case.
My soon to be 2 euro cents
December 13th, 2010, 07:10 PM
Kaspar: >> -3dB does NOT make less grain, it just limits the highlights never reaching 100% so it actually just clips dynamic range
Are you sure? How can decreased gain value work as a limiter?
Also, I do remember doing a (not scientific) test a year ago when I first got my SI2K, and it appeared that -3dB gain setting was a bit cleaner than 0dB. I'm referring to the amount of digital noise. Don't know what "grain" corresponds to in digital realm.
It'd be great if Ari chimed in on this.
December 13th, 2010, 11:15 PM
Thanks for your advice.
As a user of DVR2.0, could you tell me if it records only to ".mov" files. I could not find ".avi" files in the beta demo version.
To everyone, another question.
I tried fitting a SSD into a dataport ( digital magazine ). The P+S recorder unit does not see the SSD. My XP computer does see it in the dataport. I have a dataport dock on the computer.
I then initialised and formatted the SSD in my XP computer to NTFS. I copied a project file and camera file across to it, then put it back into the P+S recorder unit. The P+S recorder unit still does not see the SSD.
Any advice on what I am doing wrong will be greatly appreciated.
December 14th, 2010, 02:33 AM
1. The gain/grain issue - this is analogue gain - the actual voltage on the sensor is changed. So this does not work as as it does on majority of video cameras and even RED and Arri, where the digital data is scaled. Because all sensors are linear that means 1/2 of the bits are used to see the brightest stop of the image, the next stop is 1/2 of the remaining bits and so forth. The noise in all of it is uniform. So when you start to gamma the image from linear you will gain the lower stops by power function therefore you scale up the noise. So the more underexposed your image is, the more noise you will manage to "pull up".
Now there is different base noise is the sensor but when image is properly exposed you still see very little of it.
About -3DB just do a simple test, shoot something overexposed and measure the digital value of the pixel. Because the voltage is lower the pixel will never achieve full saturation but it will be flat grey, not more image information.
For shooting in really dark circumstances it helps to actually gain the camera up to get cleaner image. For example the Werner Herzog's "Cave of Forgotten Dreams" everything inside the cave is shot with +6db because we had only 3 led light-panels that were allowed to be used.
2. DVR is MOV only, not a problem fro me because I have used MOV as long as it has been an option.
3. I dont know almost anything about PS bodies, I have built my own recorders mostly - yet this year I was working on feature in Argentina that had hired one - I really was not impressed, it was v1 and had only USB connection the HDD. It didn't handle humidity as well, even worse than Arri D20.
December 14th, 2010, 08:42 AM
Thanks for the info.
1. Thanks for the info on -3db. Gavan O'Sullivan, who DPed was metering with the spotmeter and colour temp meter he normally uses on film.
Also, as I mentioned, he treated the SI2K as a film camera. His exposures were a bit hotter than I have been using but they all look better than my own, which verifies your suggestions.
2. I am a bit sad about losing native .avi files in DVR2.0. It means going through a conversion step for work on Windows PCs instead of editing from the camera files. This direct workflow was the competitive attraction we had over the RED for Windows PC.
I concede more clients asking for the SI2K were editing on FCP or AVID than on PCs, so native camera-original .mov files which can be pulled straight into FCP are likely more competitve.
I guess SI and P+S must have researched and found more work is done on macs with .mov files than with .avi files on Windows PCs. We may shoot on DVR1.681 for clients who are on WindowsPC and use DVR2.0 for Mac clients.
3. We have the later version of the P+S camera bodies. These have SATA to the D:\> drive.
As for laptop computers, we have not had a lot of luck yet getting them to work reliably as recorders. We are still using DVR1.681. What laptop have you found to be reliable in real-world conditions?
December 14th, 2010, 09:56 AM
Kaspar, what you said about SI2K noise. What is the bottom line. Do you recommend shooting at 0dB setting, because you say this does not increase noise but provides correct gamma/latitude throughout the brightness range?
What noise are we talking about - does your 0dB example include noise in the dark areas of the image, which you say won't be affected negatively by increasing gain from -3dB to 0dB?
Bob, mov's should not be an issue on PCs. What are you guys editing with. Adobe Premiere CS5 takes virtually anything and converts it into its own intermediate format, internally, on-the-fly. That includes mov's. I did not work with Cineform mov's in CS5, but all others were imported just fine.
Regarding external drive, while SSDs seem attractive, they don't seem to work reliably as external storage with SI2K/SiliconDVR. Both I and Justin Lovell reported the same issues (we use custom laptops, actually): SSD drive would randomly get data corrupted. In my case, audio started lagging behind the video - a lot, and variably throughout short (10min) clips! This never happens with HDDs, so just stay with them.
HDD vs SSD: latter wins on speed of data transfer, and because it is not susceptible to vibrations and thus can be used in rough environments like running with cam etc.
However, all CFHD needs (due to low data bandwidth) is a 5400rpm drive, so speed is not an issue here. And with the second advantage, you'd have to have your OS drive to be SSD as well - which apparently you don't.
Why SSD then?
December 14th, 2010, 09:33 PM
"What are you guys editing with?" -- Bob sheepishly looks away and replies very quietly so no one else can overhear, "Premiere Pro 2/Prospect2KV3".
"Why SSD then?" -- Bob sheepishly looks away in the other direction - "Thought I would get ablity to record longer bursts of uncompressed when needed but mainly to be able to switch out of adaptive mode".
If somebody had cobbled together a RAID array out of 50 CF cards in a device which looked like a dog comb, I probably would have bought that too.
December 14th, 2010, 09:43 PM
LOL on the dog comb :)))
Adaptive vs FS1: seriously, even the cheapest HDD with 5400rpm will let you record FS1.
If you can't, most likely you have a processor/memory/FSB bottleneck, and not the speed of the hard drive used...
Uncompressed: I toyed with it years ago, and can honestly say it won't make much difference vs CFHD FilmScan1 (called Quality 3 in SiliconDVR v1.x) on anything, including green screen.
Now. What's up with all that sheepish stuff. Go get'em, tiger!! :)
December 14th, 2010, 10:44 PM
"Sheepish stuff." - I guess it is safe to admit that I was so cheap that I tried to live with Premiere Pro 1/Cineform on a 1.6Ghz Pentium for as long as I could. It probably was not meant to work that way but it did - albeit slowly.
December 16th, 2010, 03:56 AM
A 1.6 Pentium and an SI2K? Talk about a five dollar saddle on a thoroughbred, LOL!! Godbless!
December 16th, 2010, 08:47 AM
I think I have misinfromed you.
The Pentium 1.6 was a HDV machine for a while but not for too long. I was mixing HDV and DVCAM, converting down to DVD-Video for output back in the days of Cineform Aspect. I did some behind-the-scenes for a local no-budget from a Z1 on it. It was fearsomely difficult to use, counting frames and using the cursor to jump forward and back patiently and waiting for that slight extra pause which warned that Premiere waa about to crash, control S, then close out then reboot premiere for another session.
I have been revisiting that method wth my current computer. It used to be good with 2K footage without any effects on it but something has gone wrong with the motherboard and the system itself operates slowly on its side with the rock in the rag on the motherboard, replaced with a pencil wedged in the case but not at all upright.
I did not try SI2K on the Pentium 1.6. I do have some remaining vestiges of a reality sense. It still exists, reliable to the time I swapped the video drives out and sometimes still used for sound work.
December 16th, 2010, 03:25 PM
Personally, I never use pencils in my mobos - just sharpies.
Too much lead in pencils, if you ask me - short-o-rama!
December 17th, 2010, 05:13 AM
Do you grant an exemption for using the pencil's blunt end on the motherboard and the sharpened end against the casework?
December 17th, 2010, 01:23 PM
Nein, nein, nein, nein!!!!
June 12th, 2011, 11:03 AM
Update on greenscreen. Bob is now a bit happier. Gavan O'Sullivan updated his own kit to Windows 7 CS5 and discovered Adobe has a new keying option called Ultra.
It is very clever because it can distinguish that the two tones of green caused by the small rear side window having an approx 3/4 f-stop neutral tint are intended to be the one greenscreen.
Here is a link to a short test of badly graded footage.
YouTube - ‪BOM EXPORT TEST.wmv‬‏ (http://www.youtube.com/watch?v=7K--3SLqsrg)
So now I can go out and shoot my background plates with a little more confidence. Just have to remake the mount. My original automotive window expanding mount no longer fits because of the round shape of windows these days.
June 12th, 2011, 11:37 AM
Gavan O'Sullivan updated his own kit to Windows 7 CS5 and discovered Adobe has a new keying option called Ultra.
Hardly new! It used to be called Serious Magic Ultra. Adobe bought them out a few years ago and rolled it into CS 2 or 3.
June 13th, 2011, 10:43 AM
Not new? Hopefully Adobe has done a few tweaks in the meantime.
You're talking to a person who still lights a fire to heat hot water, drives a 21yr old VN Commodore, uses a Nokia 5110, still has a 16mm film camera in the back room and would use it too if film was affordable.
It is a wonderful thing that these software tools are now available. Who in 1985 would have dreamed of doing this stuff within hours of making the images, on a computer on a coffeetable in the corner of the loungeroom?
July 9th, 2012, 01:45 AM
After a long hiatus due to lack of good equipment and editing skills on my part and no contenders for the edit putting their hand up, Gavan O'Sullivan has been editing on a new machine and doing a preliminary grade from RAW and not the default look and it appears good.