High Definition with Elphel model 333 camera

[Odd Nydren]

Originally Posted by Andrey Filippov

Odd, next week we'll order some of the new 10349 boards (and post info) usb (with hub), ide connector (44 pin, 2mm - for 2.5")

Hub - so it's a USB host? Great!

I tried to find info on the 10349 board on the wiki...no luck - I look forward to the more info post! :)

Originally Posted by Andrey Filippov

So cutting 10347, installing some components on it and soldering it on top of 10349 will make lens interface. Actually it is possible to attach that part of the 10347 board to the 16-pin connector of the 333 camera, but you'll need to add 3.3V->5V converter (and make a custom software build).

You really make me want to buy the 333 ;) - but I'll try to hold on and wait for the 353. The main reason being the harddisk - being able to use the camera on the go without a laptop is gold to me.

Once a 353 is available - would it be possible/hard to use part of the 347 prototype board to make a lens interface? I know how to solder and read spec sheets...but I'm far away from high tech stuff. I studied electronics for three years when I was 20...but that's 15years ago ;)

Originally Posted by Andrey Filippov

Current administration in this country (where I am now) seems to follow steps of the failed soviet system so familiar to me ... :-(

The way huge corporations get more and more powerful...in the end - in my humble opinion - extreme left wing meets up with the extreme right wing. Guess they have..."mutual interests". :)

[Wayne Morellini]

Originally Posted by Odd Nydren

Extended resolution - no idea :) - sprites in the border, if I remember correctly I hooked an interrupt and then used NOP (no operation) commands to delay/sync up commands so they are done when the screen cathode ray is outside the frame...if that makes sense? :)

Yep, makes sense, interesting. If we only new the tricks back when we had the machines, we might have been more inclined to keep them around. The Commodore 128 was a good selling machine in the US for years with the Amiga around.

[Phil Stone]

35mb 'original unaltered' full frame 12fps 333 clip. Its 10secs long & recorded at 70%.

http://www.tacx-video.com/Elphel/

Phil

[Wayne Morellini]

RAW, as in 3 mpixel bayer grey scale image?

How do we play it back?

Thanks, will probably look at it tomorrow or so, god willing, after I finish a few things.

[Noah Yuan-Vogel]

great looking video. i forget how nice these look compared to every video camera. no scaling, no cheap lens, no heavy processing/sharpening, what a nice clean looking image, and nice dynamic range. i mean cut the vertical res in half and double the frame rate and you have a wide 2k cinematic image right there. how is there no rolling shutter? short exposure? I mean even the 333 technically could handle anything modern camcorders could handle. the hvx200 seems to have almost identical pixelrate limitations. 960x720@60, 1280x1080@30 ~= 42MP/s (which is what elphel 333 lists as its hardware pixel compression speed limit). wow, direct capture to hard drive? sounds wonderful

[Noah Yuan-Vogel]

plays fine for me, i dont think its real raw, my computer decodes it with mpeg-4

[Phil Stone]

Sorry thats just my own words for things confusing everyone, its original video not Raw. Its 70% quality Mjpeg that was in a ogg container. I just put it in a avi container without rendering it. It looks crisp because its 3megapixels using the entire lens. Once you start only using a small part of the lens things get obviously a bit less crisp & tougher to focus.

You need a Mjpeg codec to work with the video in say Vegas etc. if you have Virtualdub it should play in that.

Phil

[Andrey Filippov]

Originally Posted by Noah Yuan-Vogel

42MP/s (which is what elphel 333 lists as its hardware pixel compression speed limit)

42 comes as 125MHz/2/1.5
125Mhz - memory/compressor clock, I believe we'll be able to get 133 at least - memory in 353 is 6ns already, but the FPGA code should be improved.
Then = FPGA is larger, so we can runs some critical parts in parallel.
/2 comes from the compressor implementation - last stages (Huffman, bit stuffer) have to run twice faster than the input pixels to guarantee there will be no buffer overflows (currently input stages of the compressor get /process data without asking next ones if they are ready).
/1.5 - comes from Bayer -> YCbCr 4:2:0 conversion, for each 4 raw sensor pixels compressor has to process 6 (4 - Y, 1 - Cb, 1 - Cr).

Originally Posted by Noah Yuan-Vogel

wow, direct capture to hard drive? sounds wonderful

Yes, that should be possible (if you mean hard drive in the camera) with IDE available in 353.

[Phil Stone]

http://www.tacx-video.com/Elphel/vlcsnap-10453.png here is a frame from the clip. (saved as a png file from the VLC player thats recommended for being able to play just about any type of video clip).

http://www.tacx-video.com/Elphel/Rab...inal-frame.jpg A 1280x1024 frame from another clip, its the same lens set at the same focus. If your looking at the full frame compressed into a browser then you will see the difference in clarity between 3megaPixels & 1megapixels combined with more grain from the lens.

http://www.tacx-video.com/Elphel/elphelDemo.wmv a little clip made from the older 313 footage (1280x1024 sensor 22fps), please right click & select 'save target' as its got music & its got to play properly.


Phil

[Wayne Morellini]

Is there any example footage of the bayer pattern being recorded as grey scale, and of it debayered and converted to video (for playback)?

Thanks

Wayne.

[Noah Yuan-Vogel]

Seems like it would be tough to compress a bayer image, it might look really noisy to jpeg compression, and jpeg doesnt deal well with noise. I dont doubt it would mess up colors. maybe separating out each color and compressing them to jpeg separately?

[Odd Nydren]

Well that's what's proposed...separate the colours and then grayscale JPG them separately...then combine them in post.

yes - I would also very much like to see what could be done using this!

//O.

[Steven Mingam]

I can give you raw file of the four planes if you want to play with it. I would have test it myself but i don't have photoshop and the gimp raw import function is not really working...
I have done experiment with bayer compression some times ago, using DCT only on the red and blue channel, leaving the green untouched and then range coding everything. Size wasn't bad at all, but i had some bug in the quantization, so i've no idea about the quality.
Time to get back to work, i guess :)

Oh btw, i was here because i read some news about a Dirac Pro profile, even if the news is quite old, it sound very interesting, as it should be supported by the BBC...
http://sourceforge.net/forum/forum.php?forum_id=609681

[Odd Nydren]

Yes please post the file!
Do you know what kind of hardware captured the raw file?

Also - andrey? Is there a raw file from the 333 that we can play around with?

I work a lot with 3dsMax (its my day job) and I believe it would be quite easy to write a script that can separate the colours and then save them as Jpeg + load and combine...just to try out the idea.

I would of course post my findings.

//O.

[Matteo Pozzi]

amazing display of the power of the elphel camera Phil!
if with the 353 we can achieve a 2000 x 850 pixel at 24fps without exposure difference from left to right side we could make very good cinema camera!