4:4:4 12-bit Uncompressed DVX100

[Juan P. Pertierra]

duh. you're right, shake does that.

i'm not dumb, i'm just slow. :)

[Juan P. Pertierra]

Ok, so i translated the TIFF's to cineon files, Shake doesn't really let you pick between 10bit, log/lin or 12-bit so i'm assuming it's 10-bit linear. That lowered the total size to 175MB. I compressed the frames with zip and now it's 140MB, and trying to upload it! stay tuned....

Not a very exciting clip, just my cat and me in the background pressing the capture button! There is only ONE bit that has speckles, that one pesky bit in the green channel...i'll try to figure out the correct capacitor tomorrow morning.

[Randall Larsen]

Juan (and listmembers),

Will FCP handle the Cineon format? I would like to see at least 1 second of the raw tiff files (I can play them on my SGI). The point of recording raw files is to get the maximum leeway for luminance and color correction for a film transfer.

Uprezzing:
I would be willing to uprez the tiff files on my pentium but I only have a trial license of PhotoZoom. I guess that means the stream would be watermarked. I also have access to a 240 processor Linux Cluster at the University of Hawaii; however,
PhotoZoom doesn't come in a Linux flavor. Is he algorithm published? Perhaps we could write our own code and parallelize it using MPI.

Shifting gears:
The Altera docs for the Apex 20k say Cout is 9pf max if that is helpful in choosing output capacitors. Are you using a pull up resistor or is that part of the DIO unit?

Aside:
If anyone lives near Boston, San Jose, or Toronto Altera will have a free one day DSP Video seminar (June 21 in San Jose).
See altera.com for details.

[Laurence Maher]

Wow,

Juan knows what he's doing. Man, if I only had such knowledge. This is truely the most inspiring thread I've ever read as a filmmaker. You guys are all a hoot, and real go getters. I'm glad to know there are other "filmmakers" out there (not just videographers). I'm seriously interested in doing the dvx upgrade when it's ready to go. Keep me posted at actionvideo@charter.net.

[Juan P. Pertierra]

The clip is done uploading...note that this is completely uncorrected, straight from the CCD's. The alignment is also uncorrected but it looks fine to me, let me know if there are any huge problems and I can modify the code.

http://expert.cc.purdue.edu/~pertierr/output.zip

[Randall Larsen]

Juan,

Do we have to ftp this since its a zip file? I am getting "not found" error messages from your server. The Cap files downloaded just fine earlier.

http://expert.cc.purdue.edu/~pertierr/output.zip

Thanks in advance,

Randall Larsen

[Juan P. Pertierra]

My bad! uploaded it to the wrong directory...try it now!

[Randall Larsen]

Juan,

Never mind now its downloading!

Thanks

[Les Dit]

You should have used f.@@@@.cin to resave them in shake. That forces the padding to 4 digits, making them read correctly in alphabetical order. ( allowing easy compositor usage )
ie: f1.cin would be f.0001.cin

-Les

[Randall Larsen]

Bravo Juan!

I haven't transfered the files to the apple yet. However, I looked at them in Maya fcheck. I was able to see that that the pesky speckles are only on the green channel.

Juan's cat is charming. The alignment seems fairly good. I am not sure whether the ND used is optimal.

Fcheck reads the header file size as 773 by 495. Too bad the vertical lines are only 495! ProZoom should still be able to make a nice 720P image.

[Juan P. Pertierra]

Thanks Les! I will do that for all the following uploads.

Randall: the clip was shot with no ND filter on. I originally tried ND1, but it was pretty dark.

I've gotten rid of the speckles on all the other 29 ines using caps, so it seems like it's just a matter of finding the right combination to get rid of the remaining ones on that one line.

About resolution, the actual frame read out by the A/D is 495 lines, but usually there are one or two black lines at the bottom, and on top of that the vertical shake i spoke about before that i had to correct for, takes off the top vertical line every two frames...so the effective line count is around 493. I do think however that my constants are a bit off and I am loosing one vertical column on the right, because the images ARE 773 wide and there should be no black column on the left hand side...

I uprezzed it to 1240x~720P using the batch feature of photozoom, and it looks awesome. I've only seen a few clips of the JVC HD camcorder, but i'd like to see how this compares :)

If there are enough people without photozoom, I can upload an HD version of this clip, but i will have to take the original one down due to my quota. Also remember that photozoom has a miriad of options so I am basically using paramters to make it more soft than sharp...seems to work best to my eyes.

If you haven't done so, to see my uprez settings check out the cap10_RAW.tif link...I still think cap10_RAW.tif is the best results so far.

Juan

[Randall Larsen]

Juan,

There are speckles on all three channels (blue has less) when the the file is played back frame by frame (at least in fcheck). Different speckles on different frames of course.

There is also the alphabetical sequencing problem previously mentioned. F1 sequences to F10 next. The cats head turning and ear twisting is actually quite interesting.

It might be nice to put a Macbeth or CSC color chart, a gray card, and a resolution chart, in at least a frame grab.

I can't seem to scale the frames up with the zoom command in fcheck (the file pixelates). I know the resolution is there in the file. I will try to look at the files on the apple and perhaps try my hand at a color correction.

[Stephen van Vuuren]

Juan:

I am upgrading my web-site - I now have 2 GB of space of which I use about 250 MB and 30 GB of bandwidth a month of which I use around 10 GB, so I could host your clip.

[Les Dit]

There was an interesting 'photozoom' compare back in one of the JVC HDV threads. Personally I am impressed on how well the photozoom product adds high frequency detail to blowup images. I use it all the time for my digital still camera when making prints. One must realize that while it does add some pleasing details, they are mostly artificial. It takes areas where there is a defined edge and makes those edges 'squarewave' sharp. But it cannot bring back details that were lost due ti the inadequate initial sampling. So you can blow up an image to 2X size, and have edges in the result that are one pixel edge transitions, but they are fake. The subtle detail is missing.
The results are pleasing to the eye, however. Our vision system loves to grab onto sharp edges, that's how human vision works.

The funny part of the DV resolution blowups compared with the HD images was that few people wanted to compare them with the HD also blown up with photozoom! Now that would be unfair :)

It's true that blowing up uncompressed images should work better. It would be interesting to raw read the JVC ccd and get the bit depth out of that, with the extra detail as well. The Bayer pattern would have to be dealt with, but that can be post processed after capture.

Juan, did you order the Rockwell HD CMOS eval kit? I'm curious if you have the data sheets for that, and if you can post them? Is it true that the eval kit is $10K ?Have you seen that camera one small company is making with that chip? Raw output is included, but no price. Kinneta, or some such company name.

-Les

[Randall Larsen]

Les, and Listmembers, (with a Challenge to Juan)

I mentioned the Rockwell Chip recently on the list. The evaluation kit is $10,000 but the chips themselves are only
$1490.

Here is a link to the mfg. site and a brief quote of the data sheet:

Quote*******************
ProCamHD 3560 Evaluation Kit includes:


• ProCamHD CMOS visible sensor (1936x1086 pixel format, 2/3” optical format)
• PC board (video and CameraLink outputs available)
• C mount Lens: f /1.4, 12.5 mm focal length
• Lens mount and associated hardware
• Matrox HEL 1M SFCL frame grabber and MIL Lite SDK
• CameraLink cable
• USB cable (supplies power to sensor, not used for communication with PC)
• RSC ProCam control software for Windows 2000/XP
• Sensor configuration disk
• Introductory documentation

System requirements:

Customer supplied PC with PCI-X slot, Windows 2000/XP, 500MHz P3 or better.

From my inquiry:
Regarding the ProCamHD Evaluation Kits: our kits are assembled upon receipt of order. They sell for $10,000 and contain all the components required to evaluate our sensor (a ProCamHD monochrome sensor is included in that price). This price also includes limited technical support via phone during the evaluation process to ensure the customer can exercise the ProCamHD sensor to its maximum capabilities. Please allow eight (8) weeks after order submission for the assembly, certification and testing of the kit prior to shipment. I am enclosing a brief description of the kit for your file.

Budgetary pricing: (sample quantities 10-49 units):
ProCamHD 3560P $1,470 per unit

This price includes color or monochrome, with microlenses, in a ceramic 97-pin uPGA package. I would be happy to provide a formal quotation if necessary.

We do not have defective sensors available for prototype development.

We have schematics available for the camera board of the Evaluation Kit as well as a draft of the sensor datasheet (currently in final edit).

Altasens, Inc.
1049 Camino Dos Rios
Thousand Oaks, CA 91358
Phone: 805-373-4964
Fax: 805-373-4200
E-mail: dhowe@altasens.com
www.altasens.com
unquote************************************

(newly spunoff from Rockwell with some Japanese ownership).

Maybe we should all chip in and buy Juan one to three of these chips (maybe he doesn't need the full development kit).

Should Juan build a one chip (bayer color filter) or a three chip HD camera?

Juan would you like to try to do something with one of my
dead DXC-750 heads (a useable prism block)?

I have some prism blocks that might work with these chips.
One would come from a dead Sony DXC750 (a 700 line NTSC chip camera), another is from an older Ikegami plumbicon camera, another is a 35mm (big) unit from the RCA TK-41C.

The frame store is the expensive piece. If Juan's TI firewire interface can be built then the MAC Laptop or G5 performs that function.

If might be fun to build a unit that could sit in the place of the film gate in a Motion picutre camera such as my Kinor 35H. The kinor already has the optical viewfinder Cinematogs prefer.

Dalsa has high rate 35mm sized CMOS chips but they aren't releasing them because they are building their own camera. One might be able to use two of the industrial parts (20fps) and switch between them. However the rockwell part does 60fps at
720p or 1080p.

Another option would be to use several Kodak or Dalsa 14Megapixel still sensors and multiplex them to get 24fps at an unheard of resolution (close to IMAX quality after post processing).

Different sectors in the Dalsa or Kodak chip could be driven an the image put together in a frame store so perhaps it wouldn't take more than four sensors.

Juan what do you think? Would you like to build an HD camera?
Maybe it would be cheaper to just buy the forthcoming JVC 870?

However perhaps a design that was performance driven rather than business model driven would be marketable to those who really want to make Films with the gear.


Kinetta and the Thomson Viper are being discussed extensively
on several CML professional cinematography and post lists that I subscribe to. 2K resolution is an adequate substitute for film.

The bottom line is that although film has a potential resolution of 4K only about 1K is seen on the theater screen because of the Geneva movement.