WOW!
 

ok, i'm no video specialist but i am supremely amazed.

This is a (approximately) 720P HD version of the last frame i posted, up-rezzed with PhotoZoom pro(S-spline) but other than that completely uncorrected(i.e. color still looks weird). I used the 'soft' photo setting which minimizes sharpening. I have a huge worklight lighting the scene, so the edge of the silver case is clipping the CCD's in the raw footage, but in the DV footage the entire silver case, white clothing and half of the oil box is clipped.

S-Spline understandably has a lot of trouble with the speckles, because it thinks they are details and they endup looking worse, but check out the areas that are speckle-free...wow! Since the original image is a strange aspect ratio, it resized to 1240x794. Specially details like the mobil-1 logo and the edges of the TV look pretty darn near perfect.

http://expert.cc.purdue.edu/~pertierr/cap6_HD.tif

Cineon technically can store uncompressed linear 10bit data. But I don't know of a single application that would be able to open such a file. If you go that route, be prepared to have people shrug and walk away...

- ben

Hmm interesting...all the apps i use, Apple Shake, FCP and Photoshop open cineon files. So, does Adobe AfterEffects/Premiere or Vegas not handle them?

I'm only asking because I did some tests and indeed it can store the exact same uncompressed information but with 10-bit or 12-bit packed data, reducing the file size by at least 25%, yet mantaining the exact same data as a 16-bit TIFF or RAW file.

Cheers,
Juan

After Effect yes, Vegas, no (unless added in v5), Premiere no, Premiere Pro maybe.

Thanks Stephen...i guess all of this can be made an option. Cineon will keep the same quality but will allow to get more video on the drive.

Unless there is another widely used standard that allows 10/12bpc depths...i'd like to keep it an option for those who have software that can handle it.

CLIPS:
The raw 'film strips' are being uploaded, the RED channel is already uploaded. The other two should be done in under 2 hours. To open the raw files, open the file in photoshop CS as a RAW file, using the following settings:
Width:1001
Height:59900
Depth:16-bit
Byte Order:Mac
Header Size:800
Click OK, it will tell you the selected image is smaller than the file, yadda yadda(always does this), click OK again.

Remember that the blue images are flipped horizontally, and as of this post only the RED is up:

http://expert.cc.purdue.edu/~pertierr/R_OUT16.raw
http://expert.cc.purdue.edu/~pertierr/G_OUT16.raw
http://expert.cc.purdue.edu/~pertierr/B_OUT16.raw

Juan

Let me get this straight -- Shake, FCP and PS opened 10bit linear Cineons? I would be shocked to hear that they all supported an entirely unused format -- 99.999% of all Cineon files are 10bit log, since that's how they come out of the film scanner, and that's how they go to the film recorder...

The reality is that Cineon in general is just not supported in many applications, and I think it doesn't make sense for an application such as this... I still think the best route is to record raw 10/12 bit files, and provide a conversion app. That way you don't need to do any processing on the data as you write it to disk...

- ben

That's a good suggestion, Ben...just record packed 36-bit data to maximize disk usage, and then provide an app that decodes it just like i've been doing.

My current final design does everything on the fly, but i hadn't considered that if everything is decoded into 16-bit RGB files for NLE instead of 12-bit files, there is a lot of valuable unused space on the drive....hmmm.

All R,G,B raw frame strips have been uploaded at the links above.

W00T W00T!
 

O happy day.

I have gotten rid of the speckles!

Now taking suggestions for test scenes/clips.

It was indeed noise in the lines as predicted a few messages ago, and it was solved with the use of some coupling capacitors.

Juan

Great work Juan! How about posting a few noise free stills? It would be nice to see some outside shots.

Fantastic!

High-color saturated images.
High-luma contrast images.

Overexposed scene. Fill up a room with as much light as you have.

And to make it really cool to compare, the DV versions.

Don't remember if this was ever asked. This will all work at 24P, 24PA, 30P, and 60i? How does it handle this sampling rate?

Peter, in the process of putting some test frames together. Will have to wait until tomrrow to take some outdoor shots, but for now i'm just re-taking the last shot i posted without the speckles.

Yes, it will work on every mode, although there is no difference between 24P and 24PA at the RAW level, because there is no pull up/down whatsoever. you get the actual 24/30 frames.

Juan

Here's a quick and dirty(not any more) frame. It's very dark, and i'm so happy i got rid of the speckles i can't color-correct straight...so i just uploaded a PSD file with the raw R,G,B layers for you to color correct(align?) at will:

http://expert.cc.purdue.edu/~pertierr/cap7_RAW.psd

Next, i'm gonna try some different lighting and up-rezing....

If anyone was trying to download the .RAW film strips for the 3-second clip, i apologize because i went over quota and had to remove them. I will put together a clean clip with no dummy pixels...

Juan

Why are the frames your posting tilted?

Because the camera casing is still open, and that is the position that puts less stress on a flex cable which units the main body to the LCD half-shell.

Very clean image. There still are some speckles though. Not many, just a few. Must still be some noise in the lines.

So, once its closed you can post some images that are right-side-up?

Yeah, i forgot to note...since radio shack was closed, i had to borrow a couple of capacitors from my lab and I just grabbed a couple of random capacitances...

There is so much I can do with the few i have, so this is the closest i can get. I need a capacitor a bit smaller than this to get rid of all of it, but i'll get it tomorrow.

Mark,

Yes, and actually i'm going to try one right side up tomorrow if I can get it to sit right.

I figured the angle wouldn't be a problem for assessing the color and clarity of the image...

Fantastic Juan! I've been following this thread very closely. What impressed me most about your last image (the noiseless one) was the latitude it had for correction. I've had to master a lot of footage over the years and this grab was impressively forgiving when correcting for levels and white balance.

Count me in for a mod as well! I trust you'll keep it reasonable =)

Test Chart
 

If you can get it upright (or even if you can't) it would be great if you could shoot a few different test charts. I'm particularly interested in the gain in resolution that seems pretty obvious by this image and the test HD upres I did of it. If I remember Scott's book correctly, he compared two images side by side of that focus chart... one that was after DV compression and one if a person could "hypothetically" pull the image straight off the chip. Looks like he's gonna have to put out a 3rd edition, huh? Anyway, the book has some sample color charts, (and that focus chart could give us an idea of resolution if you can't find a more detailed pro chart). Record it to DV tape to so we can see the difference.

John

This is my first try and very bad, but i decided to post it anyway since i need to go back to studying.

I'm having a lot of trouble focusing with the tiny DV viewfinder. Even a small change is evident in the RAW R,G,B frames, but until i build a method for pre-viewing the raw footage in real time this will have to do.

The frames are unaligned in adition to being slightly off focus and the rez chart is not perfectly aligned since I can't see the entire frame of the CCD in the viewfinder...

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

And yes, those few red speckles will be gone tomorrow when i find the right cap size. :)

TKS Juan,
it's time to change skin and to become one ...Viper!

Focus
 

If you capture 3 seconds worth of frames while you slowly attempt to focus, you should be able to find the frame in your sequence that is most in focus. Its probably only four or five focus points in question, so just focus slowly through them on the viewfinder.

John

Outdoor shot, very dark because it was with ND2 on. completely RAW in photoshop format...push the levels/curves hard, the info is there :)

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

I've got DV counterparts of all this, in the process of transferring all the stuff...

Another outdoor shot...this one makes a bit more use of the dynamic range. There are some speckles in the sky, which will be fixed with a larger cap, haven't had time to get it.

This is completely RAW, but what semed to work for me was to collapse the layers into one, and drop down the RGB level upper margin until it meets the edge of the histogram.

I'll post the DV frame soon...according to DV, the sky, road and houses are white.

http://expert.cc.purdue.edu/~pertierr/cap9_RAW.psd

Very excellent work Juan. I've been following this post for quite some time now and am constantly impressed by your findings.

Lately I've been thinking about getting a DVX100A, but feel that your modified (Juan-ified?) DVX100 would be a much better choice. Unfortunately, the DVX100 is pretty hard to find these days, but the DVC80 (the DVX100's interlace-only, non-filmlike counterpart) is still available. Will this modification work with the DVC80 as well? And if so, will it be as capable of recording 30fps progressive and the wider color/brightness range that you're achieving?

I have a feeling everything will translate to the DVC80 just fine since it is the exact same optical and ccd setup as the DVX100, from what I understand. Since you're pulling the data straight off the ccds, then the digital capabilities of the camera shouldn't come into play... at least that's my uneducated theory.

Nicholi,

I would have to see the technical manual for the DVC80 to know for sure. However, i too think that it probably is layed out in a very similar way, and the critical aspect is actually not the CCD's but if they share the same internal circuitry.

My mod will work exactly the same on either the DVX100 or DVX100A. However, both cameras are identical up to the A/D converters so my mod will produce the same output on both, and thus there will be no advantage to having the Advanced model, unless you want to use the DV output for something.

Cheers!
Juan

Here is my 'best guess' color correction for the last capture, and the DV counterpart. The DV frame might not be exactly the same frame but it is within 24 frames of the correct one :)

http://expert.cc.purdue.edu/~pertierr/cap9_RAW.tif
http://expert.cc.purdue.edu/~pertierr/cap9_DV.tif

OMG Juan, I am at a hotel with the laptop...I just downloaded your TREES shot....it blows my mind what you can do with that image! I would have said it was trash that dark....this is amazing!

this stuff is looking so good I almost want to forget my HD camera project!

Just discovered this thread a couple of days ago, I really think it's cool what you're trying to do/doing. In regards to the noise issue, I'm not positive but I could've sworn I saw some noise in the DV footage as well, so I'm thinking that it's probably not a result of the tweaking or 4:4:4 process. I'm curious as to how the actual image capturing (sequences) is coming along, Also much like adam mentioned earlier I wonder if this can be done with any other camera (specifialy the xl1) But hey it looks awesome so far (especially the tree's raw file) And I can't wait to see how the whole thing reproduces a moving picture!

Juan:

The tree shots are quite impressive. It just makes realize how I hate the DV Codec. I've always suspected the CCD's latitude was for more than "video" was claimed and your results really "highlight" that.

Juan:

I did some analysis on the RAW Tiff on the noise.

In the TIFF, the noise is a single pixel that register either 0 or 1 in Green channel.

The adjoining pixels are perfect and have the same R & B pixel values as the noise pixel.

To my eye, it looks like an error in writing the green pixel value.

stephen:

Thanks for the encouragement! I am pretty impressed myself with the results, and these frames are part of clips, but i have a few more lines of code to write before I can post the entire clips in color frames, and i'm also juggling finals :)

About the noise, i found out what was causing it, and it just takes a larger capacitor to fix, but i haven't had time to get it today. I added a smaller capacitor than what is needed which is why there is must less noise in these pictures...the turning point was finding that the capacitor was needed :) now it's just a matter of buying the right one...

It was just ground-loop noise in the lines.

Juan

Clipped Whites
 

So I wonder why everything gets substantially brighter between what you're pulling off, Juan, and the DV encoding. Is this simply the camera's own color correction function doing this to the image. How could a person get a sense for the final look of the lighting on this mod... from this DV frame, it seems that what you're seeing in the viewfinder is very different from the frames you're pulling. And you said you even color corrected this version too... How much darker was it pulled straight off the chips?

I think the best thing about these two frames is the red in that chair... look at all those blockies in the DV clip. In your capture version it's prestine beautiful red.

John

John,

It's not that everything gets brighter, it's just that the 8-bit DV footage clips at a much lower brightness. The reason for this is that given that the DV format standard is 8-bit but the images are captured in 12-bits, the engineers at Panasonic had to decide where to sacrifice the bits that would be missing in the DV footage. They decided to sacrifice hi-end brightness, but gaining more low-level detail. They could've set the white at the saturation of the CCD's and get the same latitude i'm getting, but then they would get much less dynamic range because the 8-bits would be spread over a larger area.

You're right about the viewfinder, it shows you what the DV footage is. Even though the viewfinder is not totally useless, there will be a way to preview the RAW footage out of my device.

>And you said you even color corrected this >version too... How much darker was it pulled >traight off the chips?

The RAW, uncorrected photoshop file is at:
http://expert.cc.purdue.edu/~pertierr/cap9_RAW.psd

Hmm
 

That totally makes sense, and explains why the whites are clipped the way they are. And there's no question that you're getting amazing latitude, but mostly I was referring to the low-levels and mid tones. Compare for example the shady area inside the trees. That also appears substantially darker than your DV clip. If it was only due to the way the 8-bits in the DV codec are distributed, then wouldn't the whites be clipped, but the darks and midtones be the same, or at least an approximation. They're substantially darker to my eye.

I did a color correction of the PSD you posted, trying to match the aproximate darks and midtones, while keeping the sky completely intact... it's pretty amazing to see that kind of image, I'll tell you that. Congratulations!

John

i'm observing the cap9 imgs:
1
in the DV frame there's the "same" (or worse) noise that is present in the raw frame
???
Is this caused by the presence of your gear in the circuitery?
If not there's something wrong on your DVX!

2
the DV frame seems to be interlaced: why?

PS
I'm sorry for the dryness of my english but... you're my man!!!

About the noise in the DV frame, i just noticed it and realized that I didn't unhook some test equipment while i was capturing and thus added yet another ground loop...

Now, i have no idea why this is interlaced. It was definitely captured in 24P mode, but apparently it got interlaced when i captured it in FCP or put it into Shake to grab the frame...that's strange.

Juan

How are you getting YUV signals from the ccd's? Aren't they RGB ?

Also: look at the sustained transfer rated for hard drives, not the interface burst rates. Big difference.