New DIY HD Cinema Camera Project

Jamie Varney · July 16th, 2007, 09:41 AM

Quote:

Originally Posted by Jose A. Garcia

Good news! I've got the MT9P031 complete datasheet! I'm attaching it to this post.

Thanks a ton for the datasheet! I need some time to look it over, but I at least now know that you registers you want to manipulate are the ones I specified above, 0x005 and 0x006. Stephen is right, all you need to do is reset the board if for some reason it doesnt work right, so if I were you I would go ahead and give it a shot. If nothing else you could just guess and check at the values until you get a usable frame rate.

Jose A. Garcia · July 16th, 2007, 11:51 AM

Ok, I just received a mail from Micron. They say it's perfectly possible to run the sensor with fixed framerate. They asked me for the resolution and framerate I wanted. I supose they'll send me a configuration file for DevWare.

Sergey Zagorskiy · July 17th, 2007, 12:50 AM

Hi John,
I'm Petrovich from Sumix Corp. My message relates to movie creation and the problem with long smooth recording of video stream.
Sumix is going to release new soft allowing lossless record of bayer (raw) frames to HD at 1280x1024 with 26fps, 1600x1200 with 19 fps
Recommended to use a separate unfragmeted HDD partition.
The recording is limited by available HD space.
The recorded file is supposed to be processed with special utility, which plays and converts it into avi. The format of the recorded file is opened so one can use his own ideas on compressing etc

in case of questions: zsp@sumix.com

Best regards
Petrovich
Sumix Support team

John Wyatt · July 17th, 2007, 04:44 AM

Petrovich! I was going to email you this week after reading on here mention of an improvement to the Sumix application. Things are moving very quickly on this thread and I can hardly keep up! A problem is discovered, and a couple of solutions are found almost the same day! It's a great time to be in DIY HD at the moment after a long quiet period. I really think the eventual results possible with the various cameras will make them a practical option for serious enthusiasts at this price point. Soon we won't be discussing how to make it work but how we made films (sometimes, with a steep technical slope to climb, you can almost forget why you're doing this!). Thanks for posting here about the Sumix news: I'll be in touch.

Regards,
John
(refreshed by the progress being made; the community feeling here really works for me -- thanks everyone)

Jose A. Garcia · July 17th, 2007, 05:22 AM

It's just incredible but true. Now we can say we're able to shoot full HD and 2K progresive at 24fps with great image quality and motion feel by just spending $1200-$1500 and I'm sure it'll be cheaper in two or three months. Well, less than a month if we keep on going this fast.

I know it's not a computer independent solution and it's not as easy to set up as a prosumer camera. But its workflow is much easier and faster than any 35mm film camera. You can set it up in less than 2 minutes, then shoot and you can debayer and lossless encode in 2 or 3 minutes per raw gigabyte.

I mean, we CAN do all that already, but if we develop our own boards and software based on what we already have, we can in fact sell this camera, which was not one of the main purposes of this thread when I started it, but it's always a possibility.

Anyway, I really think if we keep on working like we are now, this project can become the very first working really low cost DIY 2K camera, and all that thanks to this great community.

Jose A. Garcia · July 17th, 2007, 12:33 PM

Ok, Micron sent me register values but for shutter delay, width and height. They work as long as I don't touch anything. If I touch exposure value (pixel integration time) everything goes back to 25-26fps.

Jamie's option is much better. I tried it and it works no matter what values I choose for exposure or shutter, but you've got the wrong numbers. I'm shooting at 2048x858 but even setting it to 2048x856 and setting those values to horizontal and vertical blanking, it goes down to 16fps. It STAYS at 16fps or 15.9 sometimes, but stable enough to sync audio. We need to find the values for 24fps and we've got our camera ready to test audio and video.

Thanks!

Jose A. Garcia · July 17th, 2007, 12:48 PM

Ok, I was playing with vertical and horizontal blanking values and I have the perfect settings to set 24fps on the sensor, the camera and the interface.

Pixel Clock: 96Mhz
Vertical Blank: 0x01F0
Horizontal Blank: 0x0015

I was just trying values, I don't know if these are the best, but display shows 24,0 on both sensor and camera, and between 23.9 and 24.1 on screen. I don't think that's enough to desync a 30-40 second shoot.

Jamie Varney · July 17th, 2007, 02:04 PM

Quote:

Originally Posted by Jose A. Garcia

Ok, I was playing with vertical and horizontal blanking values and I have the perfect settings to set 24fps on the sensor, the camera and the interface.

Pixel Clock: 96Mhz
Vertical Blank: 0x01F0
Horizontal Blank: 0x0015

I was just trying values, I don't know if these are the best, but display shows 24,0 on both sensor and camera, and between 23.9 and 24.1 on screen. I don't think that's enough to desync a 30-40 second shoot.

I am glad to see that you got it working! I am guessing that the problem with my numbers comes from the fact my equation was based on the MT9T01 and not the MT9P031. I will look at the datasheet later and see if I can derive a new equation. As a quick note, it seems that by having a larger horizontal blank time will also give you higher exposer times but increase the rolling shutter artifacts.

This is great, soon you will be to the point of building an 35mm adapter for it, I bet you will be filming before to long! How is your search for a good C-Mount lens coming?

Now I am off to start selling junk on ebay so I can start working on my own camera!

Jose A. Garcia · July 17th, 2007, 04:34 PM

I just can't wait till it's finished. If everything goes ok, I'll start building the adaptor in a couple of weeks. Untill then I have to figure out how to record RAW Bayer direct to disk in realtime. I can do it using a very simple debayer algorithm but not in raw color mode. That algorithm gives a rather blurry image. That's strange cause raw bayer is suposed to be faster than debayered, but anyway...

I've got a question. I know I'm capturing in 8bit mode, but if I'm capturing in raw bayer, is it 4:4:4, 4:2:2, 4:2:0...? I'm not too sure about that.

Thanks.

P.S. The C-Mount lens... I haven't seen much yet. Also we need an equation that can reduce rolling shutter to a minimum.

Jamie Varney · July 17th, 2007, 05:16 PM

Quote:

Originally Posted by Jose A. Garcia

I just can't wait till it's finished. If everything goes ok, I'll start building the adaptor in a couple of weeks. Untill then I have to figure out how to record RAW Bayer direct to disk in realtime. I can do it using a very simple debayer algorithm but not in raw color mode. That algorithm gives a rather blurry image. That's strange cause raw bayer is suposed to be faster than debayered, but anyway...

I've got a question. I know I'm capturing in 8bit mode, but if I'm capturing in raw bayer, is it 4:4:4, 4:2:2, 4:2:0...? I'm not too sure about that.

Thanks.

P.S. The C-Mount lens... I haven't seen much yet. Also we need an equation that can reduce rolling shutter to a minimum.

When you capture the debayered video to the hard drive what format does it save it in? Also exactly what happen when you try and save the raw bayer data to the disk, does it just drop frames or what?

I think that being a single bayer sensor the data from it would be considered the equivalent of 4:2:2 since you have twice as much green information as you have red and blue. I could be wrong about this though.

Jose A. Garcia · July 17th, 2007, 05:49 PM

When debayered, it's stored in raw debayered format. I can also choose to save it in AVI uncompressed but I need to capture to RAM to maintain 24fps.

When it cannot save to disk at full speed it just slows down and sets a lower framerate. It never drops frames.

4:2:2 is ok. It still has enough color info for later processing.

John Wyatt · July 18th, 2007, 05:59 AM

"4:2:2" is associated with the video colour model of a luminance channel ("4") and two colour channels ("2:2"). As Jamie says, it's probably accurate enough to describe the eventual RGB colour model (three colour channels) as being "4:2:2" since Bayer (one greyscale channel) has twice as much green ("4") information as red and blue ("2:2") information.

Take Vos · July 18th, 2007, 07:57 AM

John,

You could say that, but on the other hand a debayer algorithm will try and restore the full RGB4:4:4, before it is edited.

This restoring of 4:4:4 could also be done on 4:2:2 YUV images, it just hasn't been done very often. I think Natress tools can restore 4:1:1 to 4:2:2 with DV footage.

If you are editing in say Final Cut Pro in YUV4:2:2 then although you can restore to YUV4:4:4, it will revert back to YUV4:2:2 because of the underlying codec.

Cheers,
Take

John Wyatt · July 18th, 2007, 08:22 AM

Take -- that is a good point. Depending on how good the deBayer is, then the full colour restoration of the eventual RGB could conservatively be called "better than 4:2:2". How close you get to theoretical "4:4:4" I guess comes down to the quality of the algorithm used. I suppose that subjectively, if it looks good, it is good. Being independents, we aught to call it "4:3:3" !
John.

Jose A. Garcia · July 18th, 2007, 09:49 AM

That sounds good. And as we're debayering in post, we can have any algorithm out there, or modify one to get even more quality.

July 17th, 2007, 12:50 AM	#168
Sergey Zagorskiy Tourist Join Date: Jul 2007 Location: Ukraine Posts: 1	Hello from Sumix Hi John, I'm Petrovich from Sumix Corp. My message relates to movie creation and the problem with long smooth recording of video stream. Sumix is going to release new soft allowing lossless record of bayer (raw) frames to HD at 1280x1024 with 26fps, 1600x1200 with 19 fps Recommended to use a separate unfragmeted HDD partition. The recording is limited by available HD space. The recorded file is supposed to be processed with special utility, which plays and converts it into avi. The format of the recorded file is opened so one can use his own ideas on compressing etc in case of questions: zsp@sumix.com Best regards Petrovich Sumix Support team

July 18th, 2007, 07:57 AM	#178
Take Vos Major Player Join Date: Mar 2007 Location: Amsterdam The Netherlands Posts: 200	John, You could say that, but on the other hand a debayer algorithm will try and restore the full RGB4:4:4, before it is edited. This restoring of 4:4:4 could also be done on 4:2:2 YUV images, it just hasn't been done very often. I think Natress tools can restore 4:1:1 to 4:2:2 with DV footage. If you are editing in say Final Cut Pro in YUV4:2:2 then although you can restore to YUV4:4:4, it will revert back to YUV4:2:2 because of the underlying codec. Cheers, Take __________________ VOSGAMES, http://www.vosgames.nl/ developer of Boom Recorder and Mirage Recorder

July 16th, 2007, 11:51 AM	#167
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	Ok, I just received a mail from Micron. They say it's perfectly possible to run the sensor with fixed framerate. They asked me for the resolution and framerate I wanted. I supose they'll send me a configuration file for DevWare.

July 17th, 2007, 04:44 AM	#169
John Wyatt Regular Crew Join Date: Jan 2006 Location: West Country, UK Posts: 141	Petrovich! I was going to email you this week after reading on here mention of an improvement to the Sumix application. Things are moving very quickly on this thread and I can hardly keep up! A problem is discovered, and a couple of solutions are found almost the same day! It's a great time to be in DIY HD at the moment after a long quiet period. I really think the eventual results possible with the various cameras will make them a practical option for serious enthusiasts at this price point. Soon we won't be discussing how to make it work but how we made films (sometimes, with a steep technical slope to climb, you can almost forget why you're doing this!). Thanks for posting here about the Sumix news: I'll be in touch. Regards, John (refreshed by the progress being made; the community feeling here really works for me -- thanks everyone)

July 17th, 2007, 05:22 AM	#170
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	It's just incredible but true. Now we can say we're able to shoot full HD and 2K progresive at 24fps with great image quality and motion feel by just spending $1200-$1500 and I'm sure it'll be cheaper in two or three months. Well, less than a month if we keep on going this fast. I know it's not a computer independent solution and it's not as easy to set up as a prosumer camera. But its workflow is much easier and faster than any 35mm film camera. You can set it up in less than 2 minutes, then shoot and you can debayer and lossless encode in 2 or 3 minutes per raw gigabyte. I mean, we CAN do all that already, but if we develop our own boards and software based on what we already have, we can in fact sell this camera, which was not one of the main purposes of this thread when I started it, but it's always a possibility. Anyway, I really think if we keep on working like we are now, this project can become the very first working really low cost DIY 2K camera, and all that thanks to this great community.

July 17th, 2007, 12:33 PM	#171
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	Ok, Micron sent me register values but for shutter delay, width and height. They work as long as I don't touch anything. If I touch exposure value (pixel integration time) everything goes back to 25-26fps. Jamie's option is much better. I tried it and it works no matter what values I choose for exposure or shutter, but you've got the wrong numbers. I'm shooting at 2048x858 but even setting it to 2048x856 and setting those values to horizontal and vertical blanking, it goes down to 16fps. It STAYS at 16fps or 15.9 sometimes, but stable enough to sync audio. We need to find the values for 24fps and we've got our camera ready to test audio and video. Thanks!

July 17th, 2007, 12:48 PM	#172
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	Ok, I was playing with vertical and horizontal blanking values and I have the perfect settings to set 24fps on the sensor, the camera and the interface. Pixel Clock: 96Mhz Vertical Blank: 0x01F0 Horizontal Blank: 0x0015 I was just trying values, I don't know if these are the best, but display shows 24,0 on both sensor and camera, and between 23.9 and 24.1 on screen. I don't think that's enough to desync a 30-40 second shoot.

July 17th, 2007, 04:34 PM	#174
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	I just can't wait till it's finished. If everything goes ok, I'll start building the adaptor in a couple of weeks. Untill then I have to figure out how to record RAW Bayer direct to disk in realtime. I can do it using a very simple debayer algorithm but not in raw color mode. That algorithm gives a rather blurry image. That's strange cause raw bayer is suposed to be faster than debayered, but anyway... I've got a question. I know I'm capturing in 8bit mode, but if I'm capturing in raw bayer, is it 4:4:4, 4:2:2, 4:2:0...? I'm not too sure about that. Thanks. P.S. The C-Mount lens... I haven't seen much yet. Also we need an equation that can reduce rolling shutter to a minimum.

July 17th, 2007, 05:49 PM	#176
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	When debayered, it's stored in raw debayered format. I can also choose to save it in AVI uncompressed but I need to capture to RAM to maintain 24fps. When it cannot save to disk at full speed it just slows down and sets a lower framerate. It never drops frames. 4:2:2 is ok. It still has enough color info for later processing.

July 18th, 2007, 05:59 AM	#177
John Wyatt Regular Crew Join Date: Jan 2006 Location: West Country, UK Posts: 141	"4:2:2" is associated with the video colour model of a luminance channel ("4") and two colour channels ("2:2"). As Jamie says, it's probably accurate enough to describe the eventual RGB colour model (three colour channels) as being "4:2:2" since Bayer (one greyscale channel) has twice as much green ("4") information as red and blue ("2:2") information.

July 18th, 2007, 08:22 AM	#179
John Wyatt Regular Crew Join Date: Jan 2006 Location: West Country, UK Posts: 141	Take -- that is a good point. Depending on how good the deBayer is, then the full colour restoration of the eventual RGB could conservatively be called "better than 4:2:2". How close you get to theoretical "4:4:4" I guess comes down to the quality of the algorithm used. I suppose that subjectively, if it looks good, it is good. Being independents, we aught to call it "4:3:3" ! John.

July 18th, 2007, 09:49 AM	#180
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	That sounds good. And as we're debayering in post, we can have any algorithm out there, or modify one to get even more quality.