I was playing around with edge enhancement on samples Juan has posted so far and I have to say, this stuff has got some juice in it! The uncompressed images can literally be sharpened into oblivion after uprezzing and they still look great. I posted a sample of Juan's cat for your viewing pleasure here:

http://www.hot.ee/whiteroom/sharpening_comparison.psd

This is a 3-layer .psd file. I uprezzed the original image to hi-def, but cropped it in the end to keep the file smaller. Comparison to original size included. Mind you, this is an old example, so pay no attention to the speckles. You know they'll be gone in the final version.

Cheers,
Chris

Updates....
 

I don't remember if I posted this or not:

http://web.ics.purdue.edu/~pertierr/cap11_HD.tif

I showed it to a senior proffessor who does a lot of 4k graphics and HD work. he was very impressed with it, he couldn't beleive it came from a DVX100. He has a macbeth chart and is interested in doing some proper green screen tests in the next few weeks.

There have been several changes, but we are now ready to assemble a final production version of the device. We've done our best to keep costs down for the initial version. Some features such as SDI are a bit costy to implement, so if there is a great demand for it, we will provide a new version or add-on module.

The final design uses USB2.0 instead of Firewire800. It will still work directly to a drive, but USB2.0 allows for a smaller device(and thus compatibility with the Mini35), and it also allows for the use of a firewire800 drive when hooked up to a laptop. Before, the firewire800 port would be occupied with the device. it makes more sense to use USB2.0 because it has enough bandwidth to handle the full quality data. Drivers will be provided so you can use your laptop as a preview screen while you record. The device also has a VGA output port so it can be hooked to any screen which supports 800x600 72Hz. We are using a tiny flat panel from a POS system.

Juan

so when can we see the web site.

Out of curiosity Juan, what is the maxium resolution you are getting from the DVX100?

773x494 pixels per frame, 36-bits of color information per pixel.

Only 493, would the PAL version get 540?

I was wondering, like the Canons and new Sony, is it possible to offset the CCD's to get interpolated 854*480 picture to match the widescreen flat TV's (using anamorphic adaptor)? Or further to archieve interpolated square pixel resolution equivalent to widesceen 2.39:1 format (what would that be 480*1148, 540*1291 PAL)? Once it went through upscaling film conversion it would be perfect for cinema production.

Without modification, you can't move the CCD's for a specific configuration. However, in the DVX the CCD's are already offset, i beleive specifically the green sensor is offset somewhat. My software doesn't yet take this into consideration, but it is very probably that I can get an even better HD image by somehow using the pixel shift.

Juan

Juan, is it possible to flip the image inside the mod? Many people are building their own Mini35 systems and these usually project a flipped image on the CCD. I am building one of my own for steadycam work, because the PSTechnik adaptor is just too heavy.

Pretty please with sugar on top,
Chris

Yup, it already does that.

Juan... many moons ago you toyed with the idea of putting a cooling device (a'la astronomy) behind the CCDs to reduce thermal noise.

Are you still toying with that idea in the future or do you have your hands full w/too many other things?

Hi,

Depending on the mechanical construction, it could be as simple as strapping a peltier stack on the back of the block carrier and cooling it. Noise would be an issue, as would battery consumption. However, this may be worth pursuing, particularly in light of the poor low light performance with this modification - well, no poorer than it ever was, but you get my drift. I'll go out on a limb and bet that at least the two LSB of that 12-bit data is noise, and probably three or four.

Phil

Hello there !
Sorry Juan and everyone, but I am way too lazy to search the whole thread
in order to find out if DVI is still in or not.
I wish it is, cause those new Apple displays sure are tempting monitoring creatures.
Keep it up
Emmanuel

If you cool the CCD's be carefull of condesation build up on the cooled surfaces. PC's were having that problem with their peltier coolers, the water would short out parts of the board, I don't know how they over came it.

<<<-- Originally posted by Juan P. Pertierra : 773x494 pixels per frame, 36-bits of color information per pixel. -->>>

At that resolution, if my math is right, you're look at about 1.6 megs per frame. Even at 24p, that's nearly 40 megabytes per second.

Have you actually found any USB2 drives that can actually sustain that? Even the best USB2 drives I've seen have only sustained about half that in benchmarks.

Has there actually been any testing of "full motion" capturing instead of single frames?

I have done full motion capture directly to a Western Digital IDE drive, i listed the model number here before but it was nothing fancy, i bought it at circuit city.

Whether a specific USB2.0 hard drive enclosure supports the maximum speed of USB2.0 that remains to be seen. My LaCie drive works fine.

Juan

Hi,

It's probably worth noting here that a filesystem implemented in firmware doesn't suffer from the same problems as one that's part of an operating system - mainly that it's one of a large number of tasks, each of which is trying to claim bus and processor time. This filesystem doesn't even need to know about directories, only needs to know how to write one kind of file, and many other shortcuts which will increase its speed.

That said, I'm fairly surprised that a standard IDE drive will write 40Mb/sec, even in raw sectors.

Phil

A quick question to clear something up, if you don't mind, Juan: when you say you're using USB 2.0, is that as a camera-to-drive interface, or drive-to-computer? Or both?

I was concerned (like I could afford this setup in the first place) that USB 2.0 wouldn't be fast enough to capture directly from the camera without dropping frames. Capturing regular DV footage from my camera directly to a USB 2 LaCie drive started dropping frames if I so much as opened a window...though I suppose there are other factors at play in my situation.

What exactly will this USB connection be used for?

Over at the 10-bit 4:4:4 and HD cinema camera threads they have discovered drives that do 50MB/s sustained and even one that does 72MB/s. Problem is I can';t remember what they are (and have to go out now), but they're there. If you want to do a search, I think Juan M. M. Fiebelkorn mentioned them about a month or two back, and they were Western Digital. These are likely to be server drives.

USB2.0 is fast enough to transfer all the data. The bottleneck is when you introduce software based systems such as a computer. Phil makes the point about having only hardware interaction.

For example, my desktop computer is about 2 years old. If I tried to capture the full data in real time using standard functions i'd run into trouble. However, the experimental digital capture card I'm using has specialized drivers which use DMA and use a direct path between the PCI Bus and the hard drives. I can capture full motion up the point where my 120GB drive is full with no problems whatsoever.

About the pixel shift...it's not really a 'shift' of the CCD's per-se. It rather seems that the green CCD is placed farther away from the prism by a tiny amount such that there is a larger projected image on the chip, and the optical information that lies between elements in the other CCD's lies ON elements in the green sensor. Afaik the sensors are not moved in the x/y direction willingly.

And no, the HD images I have posted do not take advantage of the pixel shift. I actually 'eliminated it' by resizing the images to match and aligning them which will actually yield worse detail. I'm still not sure about what procedure to use to take advantage of the shift....it seems like I would have to up-rez R,G,B channels separately in a bicubic manner, align them and THEN apply S-spline.

Anybody have a suggestion?

Juan

Use the same demosaiking algorithms used for single sensor cameras.
Just need some little modifications to work with that structure.
The drives Wayne is talking about are the WD Raptor.
They are not server disks.
SATA interface, 10,000 RPM, 72 MB/s.
Two flavors , 36 and 74 GB.
Around $200 (74GB).

keep in mind that this a total guess, but... maybe in order to take advantage of pixel shift, you would have to be converting to YUV, like the DSP does (at least i assume it does). if the R, G, and B channels are somehow converted to YUV individually first, and then the Y (luminance) channel of the original G channel is combined with the Y channel of the other two, then maybe it somehow improves image detail. or maybe it's the U and V channels from the G channel that somehow are combined with the other U, V channels, though it seems like that wouldn't make sense. but regardless, i would assume that the utilization of some other color space is a part of the whole pixel shift technology thing.

i read somewhere that the Arriscan film scanner uses pixel shift to obtain a 6k frame using a 3k scanner. maybe someone science-brained can look up the patent info (if that's even possible).

juan, can you post a new raw frame straight from the camera? i have no idea if any of the other frames i have from you have had the green channel adjusted or not.

i did a quick search and found these...

http://www.cse.ogi.edu/~linchuan/demosaic/
http://www.cse.ogi.edu/~linchuan/demosaic/adobe-raw.php
http://www-ise.stanford.edu/~tingchen/main.htm

i have no idea if those sites help any, but i know they make my head hurt. maybe someone can write one of those authors a charming email and ask for advice.

Ok I Tried a search but with no result, so anybody remembers if DVI is in or out ?
I think this could be quiet important.
Take Care
Emmanuel

I forgot a little thing....
I guess that kind of setup would require the use of an anamorphic adapter unless you want to resize the image to anything near the 4:3 aspect ratio after the interpolation...
Cause if the CCD is 794x4XX you will end with 1588x4XX (anything around a 3.3 ratio)

DVI is normally output from computers and input on veiwing equipment. I thought I found an AGP 3D card with input, but apparently was wrong on that. We discussed this over at the cinema camera technical thread (I think) and I did post a DVI input board there.

Juan P, thanks for your answers.

If you look at the cinema camera thread (and the obsura cam wiki) you'll get an idea of just how much capture performance you can squeeze out of a computer in HD, which is enough. USB2.0 has a lot more cpu performance hit than IDE and Firewire), and Rob Scott is getting much better capture peformance than expected as he learns more.


Thanks

Wayne.

Yeah, I gotcha. It's the middleman that causes such problems.

Thanks for the explanation!

Juan(M.):
Remmeber that the DVX's green channel is enlarged in both dimensions, so there is pixel shift in both axes...it's not only shifted horizontally but vertically as well...so the final aspect ratio should be about the same(?)

didn't know that, sorry.
Well, if that is the case, just two possibilities are left.
I could be an amazing high resolution image, or it can be really a mess :)

Is this for real?
 

I have been reading this forum with great interest. Juan's work looks very cool. However, after waiting for over a month for the website to appear I am begining to wonder if this thing will ever come to fruition.

Any ETA as to when the site will launch?

Thanks

Please do realize that this is essentially a one-man operation. Opening a company of any size is a huge task, both time-consuming and monetarily. Not to mention the rest of the tasks such as actually BUILDING this.

As of right now I am working on assembling final production hardware.

Juan

Also, thanks to everyone who posted info on the mosaic approach, it seems that this will indeed work, and I am modifying my code right now to do processing based on this. I think I can get a GREAT ~720P image with this system, while still mantaining the same bandwidth.....very cool. I say ~720P because with the anamorphic adapter the image will be more like 1.83 aspect ratio.

Juan

Juan,

Sorry be so eager to see something. I'm sure the more work you put into it, the better it will be.

Waiting patiently. Sorta.

Hey, that's ok...i just want to make clear that i'm not just sitting here idly. There's stuff going on which is critical to doing this correctly and legally. :)

What was the format again?
 

Hi all. I haven't posted in a while. I forget - what's the file format that Juan's mod records in? And what software can edit it? For us film out guys, Photoshop isn't a viable nonlinear option. Thanks in advance =)

Right now it outputs TIFF's, but other formats can be included if needed. For example I have code working for cineon/dpx files as well.

TIFF's can be loaded into FCP as frames if your system is fast enough, or you can create an EDL using DV footage which was recorded while recording in RAW.

Juan

Combustion 3 from Discreet could be a good editor to use if you are into visual effects work or just high end color control. Combustion 3 has a basic editor built in while supporting uncompressed files in a 16 bit color space. This will allow you to do some really high quality color correction and compositing effects while also editing your project. There are two codecs out there that I know of right now that will allow 16 bit quicktime files to be created. You could convert your captured tiff sequences along with audio tracks to one of these codecs. You could then have full FX, editing, and audio inside of Combustion 3. You can also just load the Tiff sequences into Combustion 3 if you would like. I am not sure if Combustion 3 will load EDL files. Does anybody know if this works? If it does load EDL files or another type of editing project file then you could edit your dv version of the video in any program and export a edl file. The edl file could then load your actual raw tiff video into Combustion 3 once the editing is done for color timing/correction and effects work.

Editing
 

Can we get a video file sample so we can test which editors are best suited for editing, correction, speed, ability to accuruately monitor the image, etc?

Thanks.

Yes that would be a good idea. I use After Effects with the Color Finess plugin (well its now built in - vs 6.5) and would hate to pay for combustion too.

based off what i know about juan's mod, it seems like there is a very simple and relatively user-friendly solution to the editing/finishing paradigm...

edit offline (in 23.98 fps) using the 24pa, 4:1:1, 25Mbps footage that is laid to tape. this of course will be hassle-free, as long as you use a slate or something for synching later on. then once you have picture lock, you take note of the used takes/shots. locate those tiff sequences and load them into 16-bit software, such as aftereffects or combustion. batch render them out to 23.98 fps movies, using a 16-bit lossless codec.

import those new 16-bit movies into your same FCP/avid project. don't worry that the image size is bigger than 720x480. then manually synch them up to their DV shot equivalent. this labor of synching is going to be necessary if you intend to use the camera audio, anyway (or you could greatly simplify and quicken the whole synching process by making sure to capture your DV tape clips with the frame of the slate making visual contact as your first frame, then when you import the tiff sequences into AE/combustion, just trim the clips so that this is the first frame of the rendered 16-bit clips, then open your FCP/avid project and simply reconnect/replace the DV clips in the timeline to the new 16-bit movies).

then from FCP/avid, you would create an EDL (using automatic duck if you're using FCP) to port the complete edit into aftereffects or combustion. open it up and your final edit is neatly laid out for you, with each individual trimmed shot as its own layer, ready for color grading.

trying to edit the actual 16-bit footage is going to either require one helluva expensive workstation for realtime playback, or significantly hinder your ability to edit creatively and quickly. editing offline-to-online is a lot less headachey than it used to be.

hope this helps,
jaan

Juan,

Since most of the comments on how to edit the footage (and they make sense) is to edit the dv footage and then sync up the uncompressed files, would it not make sense to have a timecode-in hookup on the mod for SMPTE timecode? For film out use, it's pretty normal to record tracks of audio on another device that's receiving the same time code as the camera (sourced from a timecode transmitter) and then match the timecodes in post when you're onlining the production. Would this add too much to the cost of the mod?