New DIY HD Cinema Camera Project

[Arnie Schlissel]

Dan, are you aware of the Elphel cameras & the Apertus opens source cinema project that's affiliated with it? I think that Elphel is currently using an Aptina sensor. The camera is open hardware with open source software. Apertus is trying to develop the software along with an open hardware controller.

Elphel, Inc. | Imaging solutions with Free software and open hardware
Home page | Apertus - Open Source Cinema

[Dan Hudgins]

Quote: [Graeme Sutherland, 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.]

The new software I plan to release soon already converts True RAW sensor data to DNG, and also does bad pixel removal (interpolation) and FPN subtraction.

The main problem with DNG is Tag 50721 XYZ to RGB color matrix. There does not seem to be any document that tells how to find the right XYZ values based on shooting test charts. It is possable to guess the "right" matrix values, but the results from DNG conversion are not good.

Direct conversion to 16bpc DPX and TIF files is a MUCH better option than doing anything with DNG files for output.

I also have DNG input, but use it as RAW data and ignore the tags that just muck up the image processing and use my own "film-like" processing video path that seems to be producing results better than what I have seen posted from DNG conversion in other programs. I sent some samples to Acam dII of their sample DNG data and asked for permission to show my version of their images but have not heard anything back from them. When my program is out you may be able to use it to process the Acam dII sample DNG files and see for yourself if any other DNG convertor does a better job, if you find one let me know I would like to look at what you think looks better about their conversion. My processing is full manual adjustment, which can help avoid issues with pink highlights and such.

If anyone has their own Acam dII (or any other True RAW cinema camera) and wants me to do a sample conversion for them they can contact me.

The main problem with recording on a computer is the speed of the SSD, and the glitches caused by the OS and hardware being interrupted. An all hardware solution using CF card array would make a much more compact camera. The software support I have developed solves all of the image processing issues once you save the true sensor data without glitches.

Quote: [Arnie Schlissel, Dan, are you aware of the Elphel cameras & the Apertus...]

I spoke to them on their forum about adding a True RAW 12bit port to their 8bit compressed camera so that the full 12bit sensor data could be extracted for recording on an external recorder using DRAM. It seems they were not interested in talking with me about what would be needed for producing a True RAW recording camera. Major camera makers are going to 12 and 14 bits, their inisting that 8 bits is "all that is needed" or whatever they say is non productive as far as where my projects are headed.

More than 8 bits are needed just to hide all banding in Digital Images, and if you are going to grade with heavy ISO boost and S-curves then there is no other result than histogram gaps if you start with just 8 bits. 8 bit linear data will not even gamma correct to 2.2 without gaps opening in the shadow areas.

Also it is not a good idea to try to de-Bayer compressed data, and any claim that the compression is 100% lossless (no change to the sensor data ADC values) at compression ratios of 10:1 or more (compaired to the packed sensor data bits or delta encoding) is just out of the picture.

If they want to take all the compression "stuff" out of their design and record the actual sensor data as is 100% un-altered, and save it as is to a file named RAW, in numbered frame files like,

00000000.RAW
00000001.RAW
00000002.RAW

Then all that is needed after I release my new de-Bayer program is to enter the RAW setup values for un-supported cameras in the manual setup prompt with values like:

X pixel size
Y pixel size
Data format, 8, 12 (two types supported now), or 16 bits
Bytes per pixel, 1, 1.5, 2
Shift Left for MSB aligment

and so on.

There are prompts for X and Y mirror and Bayer order as some sensors use various starting color pixel.

I just added support for SUMIX (tm) cameras, but when I asked SUMIX (tm) for some sample True RAW sensor data samples shot under calibrated lighting they did not reply as far as I know. Someone sent me a sample that had: gamma correction and clipping, white balance done badly, median filter noise reduction, and such done to the data, so there is nothing I can do with that to make a Preset for that camera yet, someone would need to send me a sample of actual RAW sensor data, not Bayer data that has been "goofed" with and so is unusable for extracting my idea of quality results.

==

I was talking to Jose A. Garcia about the camera development, has anyone heard from him lately?

[Finn Yarbrough]

I am a Documentary shooter, so while I am always excited about these projects, the wind comes out of my sails at the talk of any kind of tethering. This is my question: Why can't we just connect a 2.5" SSD housing to these cameras to capture the data stream? Is there some way to pre-load whatever software controls are on the PC into an on-board card (and eliminate the PC entirely)?

What I am looking for is an ENG camera that uses SSD's as a capture medium, and takes advantage of the fast transfer speeds to decrease compression a little bit and increase the bitrate of the footage. Can one of the knowledgeable in this forum help me understand what is missing?

[Dan Hudgins]

I don't think there is any reason that such a camera could not be made...

For ENG use though rolling shutter issues need to be delt with, by shooting at 48 or 72fps for 24fps results, or using a sensor with a global shutter which can increase the camera cost. For tripod use where the camera does not move fast (such as for some kinds of feature film making), the rolling shutter is less of an issue at 180 degrees...

[Dan Hudgins]

The main issue that has come up with talking with various camera makers working on True RAW recording cameras is that they seem to not want to make a "shoot-then-dump" camera because the concept of a camera that has a maximum record time of 30 seconds seems limiting the market for the camera.

But for feature film production all you need is a camera that can shoot shots up to 30 seconds, as very few movies have any shots longer than 30 seconds.

So it seems the lowest cost Digital Cinema Camera would be one that is like a web cam, but has a 30 second buffer built in using DRAM since FLASH memory can wear out over time. DRAM also can have an advantage in that it may have no bad addresses, FLASH generally requires mapping the bad addresses which makes its implementation more complex.

Since all the True RAW data writes to the buffer, the link to download the data in non-real-time can be slower than the ADC output, so even USB/LAN can be used for 2.5K data.

A net-book can read a sub-sample of the data flow from the sensor to the buffer as the monitor display, then download the data after the buffer is full or a shot is finished, passing that data on to an external TB HDD plugged into another USB port etc.

Without a buffer between the sensor and the downloading computer, it is very hard to record on a "light weight" computer full uncompressed sensor data.

After having worked with uncompressed sensor data in the de-Bayer program I am developing, I think its better than various compressed data formats in that there are NO extra artifacts added, and its not that hard to deal with the time and cost of working with uncompressed DI now that TB disks are falling in price so much and second hand computers for the "render farm" are fast enough to get the work done in less than a year.

If you were cutting the picture (workprint and then negative conform) and sound (multi-track magfilm) yourself it would probably take longer than making an uncompressed DI at home, and the DI will probably give you better results in the end since you would not have as much dust and scratches etc. It still takes time to sort through the thousands of shots and sound takes and get things right no matter if you are working with film or in digital frames, at least if you want the best results you can get DIY.

[Cole McDonald]

But for feature film production all you need is a camera that can shoot shots up to 30 seconds, as very few movies have any shots longer than 30 seconds.

Many of the 30 second shots are pulled from much longer shots when editing though... I would pass on this camera based on this one thing due to the fact that I couldn't get enough coverage for editing.

It's just an opinion, so don't take my editorial as law, other people shoot differently... but this is standard practice from what I know of Hollywood shooting, so it may impact how you develop the camera's recording system.

[Dan Hudgins]

I under stand the issue, but with a story board you can interupt the "longer" take into the parts that will actually be used.

As a micro-budget filmmaker I am faced with getting something that in the end looks like a major studio production but with tools and methods that cut out all the fat.

The idea is to make a buffered camera that can sell for about $250 and can shoot 2.5K 2.35:1 ratio and give good enough results to transfer on a DIY film recorder to 35mm color print stock, or to make a DCP with the CCed 2048x858 DPX or TIF frames.

I have written the software to do most of that so far, I have a few improvements to make to the DIY film recorder software DANCINEL.EXE (tm), but its usable as is and the results of our tests watching the 35mm film projected look usable (it requires making more than one pass in the exposure to get better than 8 bit monitor limits onto the film stock).

If I am faced with shooting 35mm short ends or 35mm color print stock and paying maybe $50000 for film and processing, vs. shooting with a $250 shoot then dump camera that has double the shooting time of a spring wound 35mm Eyemo, and maybe getting better results with the $250 camera because of the freedom of dust and scratches, then I could put up with 30 seconds shot lengths.

There is no reason you cannot make the buffer large enough to record 4 minutes which is about equal to the length of a 400' 35mm film load which is standard for filmmaking. But you would need to pay more for the larger buffer memory, so its up to the filmmaker if he wants to pay for 10x the memory he needs to make a 7 second shot.

The optimum cost vs. end result comes it at about a minimum of a 30 second buffer, since very few shots in major motion pictures are longer than that, and with an auto-clapper-slate you don't need to waste frames on slating which can be done when the frames are de-Bayered.

I have a "freeish" DI system written for working with uncompressed workflows, and will have my de-Bayer and image processing program out soon hopefully, so there is support for low cost Ture RAW recording cameras in the pipeline. People can also use my system with RED, ARRI, etc. and HDSLR cameras if they like, as well as film scans. The goal is to get good results that don't look, and sound, like they were made with a cheep compressed camcorder while not spending money that is not needed to get that result.

[Brian Drysdale]

Unfortunately, you have to deal with actors and they don't switch performances on and off into neat 30 second segments. This will an even larger problem with the actors that you'll have available for an extremely low budget film that involves dialogue. It's not impossible for scenes just involving action, it's about the same as a clockwork motor on film cameras, but for dialogue I'd make 60 seconds,if not longer, as a more practical minimum figure.

[Dan Hudgins]

You got your wish, well almost, the KineRAW-S8p (tm) is not on sale yet, but I was able to shoot some test footage for development of my free de-Bayer program DANCINEC.EXE v0.04 (tm). You can view the test at this link on Vimeo,

KineRAW-S8p (tm) lens test reel #A2 w/o links on Vimeo

Its shot 2592x1104x12bit@24fps 100% uncompressed True RAW, my software removes the bad pixels and does the FPN masking. The bandwidth was 103MB/s recording to a 60GB SSD.

60000/103 = 582 seconds or 582/60 = about 9.7minutes for the maximum shot length.

With larger SSD the record time can be longer.

The camera has two hot swap slots so there may be a firmware upgrade later that could allow non-stop recording if you swap disks during shooting, but as far as I know they have not done that yet.

As the compressed images on vimeo show losses, if you want to look at some uncompressed BMP frames from that video you can email me with "DVinfo request for uncompressed frames" in the subject line so I do not delete your email as junk.

It great that after so many years Jose's vision has gotten close to being true, if anyone is in contact with him please ask him to email me again so I can tell him about the S8, as well as the S16 and S35 models in development now.

I'm not involved with Kinefinity.com sales so you would need to contact them directly for information about when the S8 would be available. I'm just interested in supporting various DNG (CinemaDNG) type cameras that record full bandwidth uncompressed True RAW Bayer data in my free de-Bayer program for use in my freeish DI/NLE/CC/MIX software DANCAD87.EXE (tm). My programs are up on my web site for download now, and I am working on some updates that I hope to have out early next year.

The video above is a collection of some de-Bayer and lens correction tests, I was also testing my 'film look" processing so the goal was not 'accurate' color as that would tend to make the results look 'camcorder' like, rather was going for a 35mm film replacement look for use in a project I hope to shoot next year for showing in movie theaters quality. I have a more developed version of my DANCINEL.EXE (tm) program for a DIY 35mm film recorder but that is not released yet, although there is the previous version up for download now. I have some things to finish and test in the new version that adds new functions and higher image quality.

The shot with the woman walking the dog was shot non-stop and then cut up, so the length of the shots in the edit is shorter than the shot length made in the camera.

Here is what I wrote on the video's Vimeo page:

Quote: [I am testing various lenses for use in film-making projects using the Super8 format Digital Cinema Camera called the KineRAW-S8p (tm). This camera records fully uncompressed at various Ultra-HD resolutions, this video was shot at 2592x1104x12bit@24.000fps 103MB/s to high speed SSD in the camera. The camera can dump the SSD to a notebook HDD in the camera so my setup was 60GB SSD and 500GB HDD for these lens tests. The camera can take C mount, CS mount, and D mount (regular 8mm) lenses. The image area varies with the resolution mode, some modes are about Super8 size so C mount lenses made for Super8 cameras may be used. the frames were processed in a free program called DANCINEC.EXE (tm) that is on my web site for download and "beta-testing". The workflow was DNG to TIF, the TIF were edited in my 'freeish' DI/NLE/CC/MIX software DANCAD87.EXE (tm) and output as BMP frame set letterboxed 1920x1080, that was converted into a MPEG4v2 6000kbps AVI in VirtualDub (tm) and upladed here. There is no sound on this reel, although the camera can record sync sound 48000 s/s stereo, and has an autoslate beep and SMPTE TIME CODE for sync with an external sound recorder or SMPTE slate etc. The camera used is a pre-production prototype that I am testing for use with my free de-Bayer software, and for use in my own filmmaking projects. The production cameras may be somewhat different from this test camera and I may be able to improve the results that my de-Bayer program produces as I am still experimenting with various filter adjustments and code developments. this was processed mostly with v0.04 while it was being worked on. If you have any thoughts or questions about the de-Bayer you can email me through the address at my program's web site. You can google KineRAW (tm) to read more about the camera development, I am not involved in sales of the cameras, just trying to work with them to support their and other companies DNG recroding Digital Cinema Cameras. You can google DANCAD3D to find my web site.]

In case you don't see that, what I see in this post is the vimeo player showing here, rather than just a link to the video page(?).

[Dan Hudgins]

I posted a second lens test reel on Vimeo from DNG frames shot 2592x1104x12bit@24.000fps 100% uncompressed CinemaDNG type DNG frames, here is the link,

KineRAW-S8p (tm) 2.5K lens test reel B1 on Vimeo

If you have questions you can ask on the videos page or email me with 'DVinfo user question' in the subject line of your email so I know its not junk, thanks.

Progress seems to being made on all three cameras, the S35, S16 CCD, and the S8. You can contact Kinefinity (tm) about their KineRAW (tm) camera development by email at the contact on their web site if you have an interest and questions. While the camera development has expanded past the original DIY project this thread was started for, Jose's vision of a uncompressed S8 as well as S16 and S35 camera continues to the goal of an affordable solution for filmmakers.

[John Wyatt]

Great work you're doing here, Dan. I periodically check out the Kinefinity website but there's still not much change. Any news on when the cameras will be for sale? And will the S8 model really be around $250?

Cheers.

[Dan Hudgins]

Last I heard was that the production KineRAW-S8 (tm) with HDMI 1280x720 output and DNG recording to 2.5" SSD was going for a target of about 1/4 the S35 price which has a target of about $6000-$8000, so would be about $1500-$2000 if that follows through.

Anything close to $250-$499 would be a web cam like camera that would plug into a notebook computer, the images would be the same but its not a self-contained camera that way. I've mentioned that concept to them but there is no target date for such a camera for sale. An unbuffered test camera was made some years ago, but that did not shoot at up to 25fps like the self contained one does because it did not have a large buffer, you can see my comments on such a concept camera above in this thread here on DVinfo. The general feeling is that such a low cost camera for filmmaking would not be of interest to many people because of the buffer limiting the shot length to maybe 30 to 60 seconds, even though that is more than the spring on cameras like the Bolex, Eyemo, and Filmo that are or at least were used by many filmmakers that shot on film.

Most of the shots I have been making with the KineRAW-S8p (tm) prototype are shorter than one minute, so for the kind of filmmaking I'm doing it would I think be usable, if not ideal, and you could get 2.5K just like in the sample videos I made on Vimeo, so the main difference between the self contained camera and a web cam plug in for a note book would be the maximum shot length, and being able to have full resolution playback with zoom, waveforms and histograms. The viewfinder and playback on the laptop would be more limited since the self contained camera has very fast hardwired circuits for the viewfinder that allow for playback with the 100% or 200% zoom and region of interest being able to be moved around during playback so you can check focus and exposure etc.

What are your thoughts on a self contained Digital Cinema Camera vs. a lower cost web cam like camera if the image quality is the same and the cost as gone into?

Do you think there is a market for both types of S8 format 2.5K C, CS, D mount lens camera?

You can write to them about the sales dates if you like there is an email on their web site. Last I heard they are trying to get the KineRAW-S35 (tm) into field testing by April 2012, they are having several prototypes made at the factory now and if those are working right testing could begin in a few weeks time. What I have heard so far, is that sales would begin after testing has worked out in China first, so to purchase an early one you would probably need to be in China, after that you will just have to follow along and see what and when things work out.

[Dan Hudgins]

If anyone has experience hacking the GH1 or other cameras on the sensor pin-out level, if you can get in contact with me and are willing to share your information with OptimaCine.com (sm) so they can engineer an adapter to connect the GH1 or other commercial camera their PCIe card, then I can maybe pass that contact information on so he can talk to you about the details, and in that way put a True RAW tap on the camera so that full sensor data can be recorded to SSD on a computer.

He has a 2.5K sensor board already and has written the recording software to record from that sensor board, the combo of the 2.5K sensor board and PCIe interface is about maybe $500, so maybe doing True RAW from a GH1 or HDSLR would be about the same cost.

The results from his 2.5K sensor board would be similar maybe to the Vimeo videos I made using the KineRAW-S8p (tm) cameras but his OLPF and IR-cut filters are different so they would not be an exact match, I already have support for both OptimaCine.com (sm) and KineRAW.com (sm) cameras in my free de-Bayer software that is up on my web site for free download now.

If you have any questions or think you can help with such a hack, email me and let me know, please put 'GH1 RAW hack' in the email subject line, thanks. My address is: tempnulbox (at) yahoo (dot) com

[Dan Hudgins]

Anyone with camera circuit skills that wants to help with what could turn out to be an important advance in digital movie capture can contact James at OptimaCine.com (tm) to share what information he might have about how to tap the sensor signals.

[Dan Hudgins]

Updates:

1) James at OptimaCine.com (sm) finally put an email contact on his web page. He's looking into various options for revising his camera design as I understand things.

2) Kinefinity.com (sm) is finishing initial testing of their two new KineRAW-S35 (tm) prototypes, I have been processing some test DNG for calibration of monitoring tables to use in the cameras so shooting of test footage can start before too long. As I understand things, they would get back to the KineRAW-S8 (tm) re-design after the S35 goes on sale in China.

I've got my DI system processing shots and the editing and sound mixing seem to function so far, so it could be possible to finish a feature film, if anyone finds an issue they should let me know. My de-Bayer program has a general configuration so other cameras not listed in the camera type menu, such as Aaton (tm)'s 4K camera can probably work by selecting the generic RAW or DNG camera type.