New DIY HD Cinema Camera Project

[Gottfried Hofmann]

Originally Posted by Aapo Tahkola

You can't use huffyuv
or similar algorithms (effectively) unless you perform debayering, so I think it's
probably better to work on bayerized data.

Why not adjusting huffyuv to work with bayerized data? It wouln't be too difficult and one could avoid propable patent issues...

[Juan M. M. Fiebelkorn]

you can easily modify Huffyuv to work with bayer, just assume you are feeding it a 32 bit image.You just get 4 planes of a quarter resolution, two of green, 1 red, and 1 blue.The problem is you will get limited to 8 bit per color channel and you will also need to generate new statistical Huffman tables for the new data.

[Aapo Tahkola]

And some of the prediction schemes would not work as well because physical distance between comparable pixels is higher. You also couldn't perform conversion to color difference domain(r-g, b-g) since you have no green pixel at that position. Simple debayering on those pixels might work but I doubt you'd get as good compression as with huffyuv. It might still be a good solution especially considering how simple/fast huffyuv actually is.

I'm not that worried about the possible patent issues since most aspects of it can be replaced with similar methods. Only the context prediction scheme is really important. Need to ask about that.
I got my modified version to run at 3 fps on my core duo@1.8 GHz(using one core) without mmx, sse, etc. optimizations, so it doesn't seem to be fast enough. gpgpu implementation would probably be fast enough for compression. Decompression doesn't parallelize that well so it would be slower.

Anyway, I realized there is really no point in doing compression on standalone version because modern computers can't perform real-time compression to any reasonable format anyway. That is, unless you have some other reason to cut bandwidth requirements. Cheapo hard drives can write ~70-50 MB/s (depending on head position) which would be enough. I'm wondering if it would be possible to directly drive hd with arm7 cpu and sensor. Blanking times could be used to perform activity on hd with arm7(assuming it's fast enough to react :) . It would require that ata drives can accept several sectors of data without intervention between frames.
According to http://home.nikocity.de/andymon/hfg/Alya/alya.html , 8 bit mode might not be supported on modern drives. LCD would need fpga or clpd though.

Checking resistances on mt9p401 revealed that it is likely pin-compatible with mt9p031.

[Alex Stewart]

Long time no post for me.... Since my last post I've graduated college, found a paying job, and gotten ready for my wedding... Which is this Saturday :)

And also as I'm sure some of you noticed I got my system to cough up a image !!!
The color is awful because the i2c bus is not hooked up ( is I'm using 100% default gains and white balance). I did not make much progress do to a nasty kernal panic that the ppi driver on the blackfin would kick out..... Hapiily that issue is solved now.

So!! At the moment I'm rigging the i2c bus up so I can mess with the gain, then I'll be piping my data into vlc to stream some video.

This is mainly for debugging, the bf can only encode at about 10 fps into mp4 at this res,
Once that works I'll be attaching a hdmi and IDE interface to the blackfin.
Piping in on archecture of my system, the idea is for the Fpga to do the heavy lifting ( ie decode the bayer, and convert to yuv) so the data stream from the Fpga will be raw full res yuv data, this gets piped into two adv202"s which plop out a jpg2000 frame that the DSP' dumps to storage medium .... Or streams acrosss Ethernet...

The nice thing is, if say I want to aide in compression , the dsp' has enough ummf but so foes the Fpga....

Hince the hardware, once I'm done, it's modular enough to support just about any software and cinema work flow you could think of....

( ps.... I plan to sell the boards at $600 ea)
(pps.... My genious 3d animator brother is also doing up a spiffy case I'll have milled out of billet Al. ........ Red one here I come :p )


Please forgive the horrid spelling and formatting, I'm typing this on a iPod touch since my laptop is in a box.... Somewhere.....

( sad thing is I can ping it... So I know it's near ...,.)

By or now folks !!! I'll be back with hopefully some video clips in about three weeks

:)

[Aapo Tahkola]

Waiting for some equipment so I can make the pcbs.

[Anmol Mishra]

just wondering if anyone has compared the Ibis sensor Marlin F-131c cameras with a HV20 to compare quality ? I'm sure its worse than the CCD MVC's but how does it compare to consumer HD camcorders ?

Thanks!

Originally Posted by Solomon Chase

The Marlin F-131C uses the ibis-5a sensor. Search the board for info. Ibis-5a is an "industrial type" sensor used by sumix, avt, silicon imaging, pixelink, etc. in some of there cameras.

Not very good color performance, but global shutter and shallow DOF are a plus.

[Dan Hudgins]

I am developing DI editing, color correction, sound mixing, and filmout to 35mm movie film software for use with DIY Digital Cinema cameras.

I was wondering if someone has a list of uncompressed cameras that can shoot RAW sensor data better than 8bits, 1280x720p or better at exactly 24fps for sync sound shooting?

The Aptina MT9P401I12STC sensor seems like a good choice for such a camera, could you summarize the current status of your project, and how others could get results usable for Digital Cinema like the results you are getting, i.e. what needs to be purchased, who do you get it from, what softwares need to be loaded and how one gets the settings right?

Anyone who has a Digital Cinema camera that is in a state that lets them shoot 1280x720p or more, 24fps, 12bit or so, RAW sensor data please e-mail me directly so we can stay in contact as I develop my post production system more, I would like to develop some extra software de-Bayer and such to support low cost Digital Cinema cameras so users/"Beta Testers" of my programs would have more low cost Digital Cinema camera options.

If you have any questions about my "freeish" software or need support using it with DIY Digital Cinema cameras feel free to e-mail me directly or through here.

Dan Hudgins
tempnulbox [at] yahoo [dot] com
The official DANCAD3D (tm) Beta Test Web site.

I may also be able to help with some suggestions for some circuit/firmware additions to aid in having your camera work well with my programs and in general for use as a Digital Cinema camera.

[Dan Hudgins]

I have been thinking that it would be useful to make a Digital Cinema camera based on the Aptina MT9P401I12STC sensor to get 2592x1102x12bit@24fps and 2048x1152x12bit@24fps RAW sensor data. If you cannot get 2592x1102 to work then 2560x1080 may.

At any rate, to overcome the limits of the USB speed, you can build a "shoot and dump" camera that has enough FLASH or DRAM to store about a minute of shooting as RAW 12bit sensor data, then after the shot is finished the data in the buffer can be downloaded through the slow USB.

What that gets you is high quality images without any problems with the computer's harddisk speed or dropped frames.

A maximum record time of one minute per shot is fine for feature filmmaking since normal feature films do not have any single shots longer than one minute, as edited in the finished film.

If you make the buffer memory swapable, you can run the camera in two modes, "shoot and dump" or "shoot and swap". If you are shooting in "shoot and swap" mode you can put the buffer card back in to download it later. Being able to also swap lets you shoot many takes rapidly, than download them later. Being able to "shoot and dump" lets you download while you are setting up for the next take. The download time would only be about three or four times the shoot time so it should work well.

Such a camera could be an open project so that various people could make improvements to the PCB circuit and layout.

I would be interested in both using such a camera to help with my Digital Cinema software development, and to support such a camera for use with the DI software I am developing.

The reason for trying for 2592x1102x12bit@24fps is to be able to down size and squeeze the images for 35mm filmout using my film recorder program to get fewer artifacts in the end result, and to use the largest sensor area possible to overcome the resolution issues of the C/CS mount lenses that can be used.

Please feel free to e-mail me about such a camera, or anything that can shoot well at 24fps true RAW sensor data at more than 8 bits and high resolution. I would like to develop working workflows for such cameras so that people can use the cameras that work well with my programs. I have tested my suggested workflow for the RED ONE (tm) Digital Cinema camera, and have a suggested workflow for the Acam dII (tm) camera on my Web site, but I will need to wait for sample DNG frame sets from Acam dII (tm) to confirm that it works or to make adjustments. At any rate, it should not be too hard to add other RAW mode camera's to the supported workflows.

Dan Hudgins
tempnulbox [at] yahoo [dot] com
The official DANCAD3D (tm) Beta Test Web site.

[Dan Hudgins]

Thanks to the DVINFO readers that have contacted me about my software and the subject of uncompressed cameras using the Aptina sensors. I expect there will be some more developments in this area, if you know of a camera project that might be useful with my DI system please let me know more so I can look into what cameras would be a good choice for doing Digital Cinema with the software I am developing to help people make motion pictures digitally... I have been doing some tests with my DANCINEL.EXE (tm) program to transfer digital frames to 35mm movie print stock, and by using my "spread gamma" feature the results look good so far shooting off a 1600x1200 LCD monitor with time exposures of about 4 seconds and longer, depending on the filters used and the f/stop of the lens. My film recorder program also supports CRT monitors up to 2048x1536x32bpp (24bpp+8bits padding, as thats the way the video boards work.)

[Dan Hudgins]

As a follow up to my post about doing some filmout tests, I found that a ROSCO #22 filter (redish orange) gives close to color balance for 35mm Eastman Kodak Vision color print film 2383 2393 stock shooting off a LCD 1600x1200 monitor. The total exposure time is about 120 to 180 seconds at T/2.8, when using my DANCINEL.EXE (tm) in three exposure "gamma spread" mode you set the animation motor delay to 45 to 60 seconds. For printing black and white on Eastman Kodak 2302 the exposure is about 4 x 3 seconds (12 seconds total) at T/8 with no filter. The main problem shooting color printstock of a monitor is the lack of RED light, I am looking into ways to deal with this issue, but for now you can use tri-color or color seperation modes if you have the time. If you want help with the gamma adjustments feel free to contact me, my email is above etc.

[Dan Hudgins]

Are there any DIY Digital Cinema Camera projects with uncompressed true RAW sensor data under-development now?

I have been working on de-Bayer software to go with my freeish DI/NLE/CC/MIX software I am working on, and if anyone has sample sensor data RAW files, I can look into adding support for their camera(s) before I release my new program, maybe this summer or so.

I have updated my "freeish" NLE's CC control screens to work better on some Windows XP Home (tm) computers that had issues with the higher resolution video modes, its up on my web site for download now. It works with frame files up to 6K or more.

Dan Hudgins
tempnulbox [at] yahoo [dot] com
The official DANCAD3D (tm) Beta Test Web site.
San Francisco, CA USA

[Dan Hudgins]

Hi,

Things are moving forward on the software and camera front, I was talking to Jose A. Garcia, but did not hear from him on my last email, has anyone heard from him in the last few months?

Thanks for any info you have, since this was his project thread I thought I would let him know about some developments, which I can't go into in public quite yet.

Dan Hudgins
tempbulbox [at] yahoo [dot] com
The official DANCAD3D (tm) Beta Test Web site.
San Francisco, CA USA

[Arnie Schlissel]

Dan, have you seen the newly announced USB3 cameras from Point Grey?

Point Grey | The World's First USB 3.0 Camera

Preliminary pricing on the 32S2 model will be about $800US and the 63S4 will be around $2000US.

[Dan Hudgins]

Hi Arnie,

Thanks for the link about the USB 3.0 camera.

I made a SSD test program and some people there with fast computers were able to record it seems 2592x1102x12bit @ 24fps using two SSD on a RAID.

There may be some glitch issues, and a 64bit OS may work better with a quad core, but it might work. With "normal" older computers it may be hard to record above about 1280x720p uncompressed.

Cabled cameras are a bit harder to use in the field, but for studio use it could work.

The de-Bayer program I have been working on for Cinema use it getting close to release.

Someone sent me a SUMIX RAW file, but is was not uncompressed "RAW" it was poorly white balanced, gamma adjusted, and noise filtered. Once the sensor data is mucked up, you cannot un-do that. It seems you have to go into the SUMIX software and check and uncheck some things before it saves actual RAW sensor data in 12bits format.

If someone has a real 12bit RAW SUMIX file, I would like to look at that data and it would help me with the presets for the de-Bayer processing. I do have the commands to break-up the SUMIX RAW into RAW frames, for shots up to 2GB long. And they gave me their 12bit packing method so I can un-pack using that, although I need real RAW sensor data to check and see if everything is working right. at 1920x1080x1.5bytes per pixel at 24fps, 2GB is, 26 seconds, so that is enough to shoot a movie as few if any shots are longer than that. With other cameras that shoot RAW frames, there is no "limit" as my program can process shots over an hour long that way.

I also have support for SI-2K (through their DNG convertor), Kinor-2K (as DNG), ACAM dII (as DNG), and some other cameras that make RAW but are not on the market yet.

My de-Bayer should be able to process any normal 8, 12, or 16 bit RAW Bayer or monochrome data to BMP 8/24bpp, DPX 10/16bpc, CIN 10bpc, or TIF 16bpc.

Anyway, if you know any other low cost cameras that can shoot RAW at 24fps, please let me know so I can look into supporting them also if I am able. When I get my de-Bayer program posted on my Web site it will be a free download for "beta testing", between now and then people can contact me about supporting cameras or doing sample conversions.

I was talking to Jose A. Garcia about making his type of camera "real" but have not heard from him for a while, but there has been progress...

[Graeme Sutherland]

The USB camera is really interesting. I'm wondering how much work it would be to push everything (processing and controls) into software, and replace all the expensive proprietary hardware with commodity parts.

If you could bundle up a simple board to convert the raw signal to (say) Adobe's Cinema DNG, control the camera, and feed a RAID system then it would be possible to build a small, lightweight rig.

However it's probably worth keeping an eye on Intel's Thunderbolt / Lightpeak technology which is even faster than USB 3.0. And if they ever release the optical version, you could have a 100 metre cable, which gives a tethered camera a lot of flexibility.

The Register has some coverage.

Intel throws Thunderbolt ? The Register
Intel: 'PC makers took the light out of Light Peak' ? The Register

And there's the Intel product page for the truly geeky...

Thunderbolt Technology