4:4:4 12-bit Uncompressed DVX100

[Randall Larsen]

Rob, Juan

Links:

Thanks. I was able to click on my links w/o a problem. I included the urls in my post instead of naming the links because many people like to know where the link is taking them.

Block Sizes relevance:

The way I usually work in Maya and Prman I have one tif or other numbered file for each frame. So there _are_ a lot of files the way I work; however, a frame at 4k resolution is good sized.

Juan's File design:
If Juan puts all the frames in one file, I may need another application to parse his file (no big deal). If he uses quicktime or mov file structure.

Xfx block sizes:
Xfs block sizes can be from 512b to 64kb. For most applications 4kb is about right. Sometimes the journal and metadata are put on another drive to speed things up. The journal is always on a different partition than the data. Terabyte databases can be addressed since the addressing is 64bit.

Using robust off the shelf software [linux, xfs] although mor complicated) may allow a more capable and reliable system in the long run.

[Randall Larsen]

Juan,

You know to evenly light the blue screen. Hopefully you have a pure monochromatic blue (sometimes difficult to find).

Suggestion:
Try to avoid reflections of blue off the bluescreen on to your foreground.

Try to avoid having the foreground cast shadows on the blue screen. Usually the foreground and the bluescreen are lit separatey. The bluescreen is lit with very flat light.

The foreground can have key and fill and backlight. Use lights with the same color temperature 3200K. Kill any flourescent lights. Try to avoid shadowing the bluescreen.

you might include hard edges and soft fury edges and glass in your foreground objects just to see how difficult it is to composite.

Try an evenly lit and a cross lit foreground.

[Juan P. Pertierra]

Ok, while it gets dark outside I have taken a frame of a resolution chart. I know it's off center, but I tried to align it such that the sides where parallel with the CCD edges as much as I could. This is intended to make R,G,B alignment as easy as possible.

Please try it out and post exactly what shifts and resizing you did in photoshop, relative to the original file. Then we can compare each other's findings.

These are completely raw, no alignment no nothing. Ignore the few speckles as usual.

http://expert.cc.purdue.edu/~pertierr/rez_RAW.psd

Are there any guidelines for testing with a resolution chart as far as camera positioning? I did this with the resolution chart about 2-3 feet from the DVX so I had to use a bit of zoom.

I noticed what seems to be a defect of the 3-ccd approach...the blue channel has a little different focus than the other two channels, probably because it is slightly farther away than the other two CCD's due to the prism design. The stuff we have to put up with. <g>

[Randall Larsen]

Juan,

The different focus for blue and red is specified in the standard for Prism blocks. Its called offset. The reason is that the lens
focuses at a different point depending on the frequency of the light.

Higher frequencies like blue are bent more and focus closer I believe. Longer wavelengths like red are bent less and focus farther back. The lens tries to compensate by using two different kinds of glass but there is always some residual differences in focus between the colors especially at the edge of the frame.

I think I posted the link to the european standard. I think the offset is something like 30 microns(different back focal lenght) for blue for a 2/3" block. The DVX100 uses 1/3" sensors.

Maybe panasonic didn't get the offset right or they are morphing the colors in their processing? The overall back focal length in air is something like 48mm (changes for 1/3", 1/2", 2/3", and 1")

I hope each camera doesn't have its own particular misregistration? I always thought ccds should have tracking adjustments but they are usually just glued to the prism block.

[Juan P. Pertierra]

I need some more bluescreen pointers.

First of all, this is what i'm using. I've got a blue posterboard, and two work lights. The only way that I have managed to not cast shadows on the bluescreen is to locate the two lights on both sides behind the key object, pointing about 45 degrees at the blue screen. The problem is that the two lights are pretty close to the blue screen so brightness pattern can be clearly seen on both sides of the bluescreen...if i move them farther away, then it casts shadows of the key object. Im using a compressed air bottle as an object for now.

I'm thinking the ideal situation would be powerful lights with diffusers, but i don't have that...any suggestions?

[Juan P. Pertierra]

Ok guys/gals, this is my first pathetic attempt. Like I said before, never done this before so i'm going just by intuition...attempting to isolate the object that is.

I have a problem with either the lighting, or the color of the board(or both) because the blue is coming out too dark, and I can't separate it from black.

Anyway, this is a scene with just the bluescreen and a bottle of those cool armor-all wipes.

What I did is I used the 'clear' feature in photoshop to get rid of the keyed selection, and see if I could get to just keep the object alone. It replaced the background with the back-color(white), and areas that were lit with the bluescreen also appear lit in white. There's also some transparency in the wipe so i thought that would make a good test.

I couldn't keep the black letters out of the selection, someone else can probably do a better job.

I need some brighter paint, and probably better home-made lights. Any suggestions from the chroma-keyers would be greatly appreciated.

http://expert.cc.purdue.edu/~pertierr/bluescreen0.tif

http://expert.cc.purdue.edu/~pertierr/bluescreen1.tif

Since i only have two lights, the key object was lit with the same lights that are lighting the background, i just moved it towards the camera until it cast no shadow.

[Laurence Maher]

The only advice I can give is to make sure you separate your subject well with backlight, giving it a rim. Whe I don't do that, my Chroma Keys look terrible. Of course, you need to have that rim motivated by the background you're gong to replace the bluescreen with. I think you'll se a noticable difference.

[Filip Kovcin]

<<<-- Originally posted by Juan P. Pertierra : ... Any suggestions from the chroma-keyers would be greatly appreciated.
-->>>


treat this as a starting point

http://www.ultimatte.com/

and you can find many *.pdfs about ultimatte solutions. you can download also demo from their site. probbably the best soft/hard ware i know now - sicnce ZBIg (rybczynski) enjoyed ultimatte and stopped his own product for cromakeying called zbig.

all the best

filip

p.s.

i cannot find really really good pdf from ultimatte (kind of " for dummies" series) which i downloaded some years ago with exellent explanations how, when and why to use chromakeying techniques. but if i find - will send it here.

p.s. 2

but maybe you should find ZBIG demo, it's available somewhere on the net. this is really amazing soft. demo will put some grid on your screen, but... if you need it for your tests - it works perfectly

[Thomas Smet]

Hey Juan don't worry about a shadow. I would just point one light at the subject and board at an angle and just leave a shadow on one side. Maybe with the other light you could point it at the ceiling or the oposite wall to bounce a small amount of light onto the dark side of the subject. Shhoting shadowless bluescreens are fine for fake or head only composite jobs but what about virtual environments? When actors stand on a bluescreen stage there are always shadows. This is just something a good keyer has to know to deal with. Besides we know this thing will work great with a perfect blue screen setup. Even DV can look good with a perfect setup. We should test setups with uneven lighting and harsh shadows.

[Thomas Smet]

that bluescreen looks pretty dark. Bring your subject and light closer to the screen to get a nice deep blue color. Don't worry about shadows.

[Juan P. Pertierra]

Ok, so I will try and find a brighter blue.

What specifically is bad about the key I posted? I know the letters shouldn't have keyed out, but other than that what are the problems I have to fix?

Another problem is that I did some rough alignment, but if the channels aren't lining up that is going to cause trouble.

Anyone tried the alignment/resizing of the rez chart?

[John Cabrera]

Juan,

I don't know if you remember but a few weeks back I identified an issue with the green channel not only being out of alignment but also having a about a milimeters worth larger FOV. That would never allow you to align the three channels unless you did a resizing to only the green channel. You were gonna do another test with some elements close to the edge of the frame so that I could test that FOV difference in the green channel and compare it to the same frame as DV. Remember we talked about you doing one this time where you didn't knock the camera. You've been so busy with interface issue that I didn't want to remind you. But now that you're diving back into this Greenscreen area I think it's an important thing for us to figure out because unless we can get those channels perfectly aligned, you'll never be able to get a professional level key out of these frames.

John

[Laurence Maher]

Hey Juan,

Were you ever able to use that anamorphic clip to your advantage at all? Just curious.

[Thomas Smet]

I played around with the rez chart a little bit trying to line up the channels. I was having a lot of trouble with the green channel. Every time I lined up one section the other side of the image would be off. If I aligned the center of the image I got very close. I never heard it mentioned but the blue channel is off as well. It appears to be 1 pixel up so it needs to be shifted down 1 pixel. I blew my image up by 400% so I would have more precision in moving 1/4 a pixel at a time. The green channel was aligned at the center by moving down and right 2 nudges which would equal 1/2 a pixel down and 1/2 a pixel right. All 3 channels were then perfectly aligned at the center. I then checked the left side of the image and everything seemed mostly ok. The right side of the image however showed a slight problem with both the green and blue channel. I have to agree that perhaps all three channels have different distortions that would make perfect aligning very hard. It seems odd that the right side of the image would be off and not the left. The amount of distortion on the right side for the green and blue channel appears to be 1/4 of a pixel or less. I can shift the green and blue channels by 1/4 of a pixel to the right and they line up better. By doing this however the center and left side of the image are now off by 1/4 a pixel.

I hope this helps

[John Cabrera]

There's no question in my mind that the three channel are different in terms of FOV or distortion or somthing... but it's odd that the DV stills don't seem to have the same problem. The best way for us to figure this out is to anylize an exact same shot in it's RAw form as well as it's DV form. Removing the letterbox from the DV would be helpful, but that may not be possible since I know you've disconnected the joystick... but if you can, it would definitly help... also do the shot indoors so that no elements like wind can change the the composition by even a millimeter... that way you don't have to search for the exact same frame from both. And make sure that you include object with detail near the sides of the frame... since that's where we'll be looking to line things up.

Thanks,
John