No announcement yet.

Little raw cam side project...

  • Filter
  • Time
  • Show
Clear All
new posts

  • Little raw cam side project...

    Hey guys.. just thought I'd share a little info about a project I'm working on while waiting for the BMC.. it certainly helps with that

    Last week I finally got the Point Grey USB3 camera head (normally for industrial use) which has a 1/2.5" Sony Exmor CMOS built in and takes lenses with C-Mount. I already started to work on my own software using their SDK before that, but of course couldn't be sure about the quality I'd get from the camera because of the lack of any demo footage/raw framegrabs.

    Over the weekend I experimented a little with capturing footage - directly to SSD the data rates are no problem, the sweet spot is seemingly 200 MB/sec: the electronics and/or sensor in the camera head can't handle more data troughput unfortunately. The 200 MB/sec is no problem at all for any decent single SSD.

    So with that I'm currently able to get a 3200 x 1800 pixel raw 12-bit frame at 24fps or when using binning (pixels are connected to groups) I can use the whole sensor for 2048 x 1080 pixel raw 12-bit at max 60fps. At 3,2K the rolling shutter is about the same as with a 5D Mk2, at 2K using binning it is (thankfully) nearly non existent (roughly on par with Red One MX)!

    The max resolution if you go with only 8 bit is 3728 x 2096 at 24fps (using almost the full sensor), but then you obviously got very little headroom for grading.
    It's quite fun to have such a little camera head on a 5m cable - reminds me very much of the SI-2K.

    The dynamic range is about 11.5 f-stops, so (of course) it's not in the BMC ballpark. But still it's such a great thing to get real raw data and without any sh*tty camera electronics or compression in the way even at 24db analog sensor gain (max) you get quite a decent picture - of course theres noise but it looks quite organic.

    Another thing I have to take care of myself that way is bad/stuck pixels, every sensor has a few of them. That's the downside to just getting a sensor "dump" basically

    Right now I'm working on the code to safe the raw data for each frame in a Cinema DNG container so that I can use the usual "high quality" tools to process the footage. And taking a noise/bad pixel "footprint". We'll see how complicated that will get - if and what kind of profiles or further meta data is required for Adobe Camera Raw etc. to do a good de-bayer.

    Keep you posted...

  • #2
    Quite Impressive - looks like, on a weekend you get as far, as the DBolex folks in a year


    • #3
      Hehe to be fair that was the easy part.. I didn't design any electronics let a lone a housing for the whole thing.. (head + embedded pc + ssd caddy) .. much left to be done...


      • #4
        This is impressive to hear about it.. interesting to hear about your progress.. Please continue!! Oh don't forget to document everything


        • #5
          Sounds great ... looking forward to the footage


          • #6
            That's awesome. Keep us posted.
            Facebook - Angelis Digital Studio


            • #7
              I demand an update!!!


              • #8
                Originally posted by mhood View Post
                I demand an update!!!


                • #9
                  Is that the Flea3 FL3-U3-88S2C-C ?

                  $945 USD for the camera head


                  Last edited by adam777; 11-28-2012, 06:02 AM.


                  • #10
                    Exactly, that's the one!

                    Still working on the cDNG stuff.. for that I also had a closer look at the publicly available DNGs from the BMCC. There's a bunch of colorimetric info in there which I guess helps Adobe Camera Raw and the other DNG readers to correctly interpret and get the most out of the raw bayer data.
                    One interesting detail is that it even employs a LUT which is present in every BMCC DNG... this maps the 12 bit (1-4096) values in the raw data to 16 bit values (1-65536) ... although I don't know how strict the readers adhere to that. I guess I'll have to do some experimentation in using a similar LUT and see if it makes any differences in the ACR output or if it's just plain ignored

                    I made a quick graph of the LUT that the BMCC writes to every of it's DNGs:

                    Interestingly this maps more than half of the possible input values to the lower 8th of the output values. There's even a bunch of the 12 bit input that's mapped to exactly the same output values.. so if this LUT would strictly be adhered to the outcome would be more 11 bit like (even a little less). Which is kinda strange. The curve looks like it's used to get back more contrast into the very flat S-Log raw data, so perhaps if this LUT is used it's just the starting point for ACR & co, and by tweaking their parameters you can get to the original flat data again.... weird
                    And also there are two entries for Black Level and White Level that instruct the reader to clip below 256 and above 60074 (from the 16 bit output value) - again if adhered to but perhaps that's just the starting point for the default settings.


                    • #11
                      When can I pre-order?


                      • #12
                        Hehe .. it's not that kinda project unfortunately... at least nothing planned yet. Have to evaluate the performance of the Sensor. But so far it looks quite nice.

                        Update: I got the first Cinema DNG recording/converting working.. so far the DNGs only work in a few programs, Adobe Camera Raw won't read them and I have yet to find out why.. something is still missing it seems :/
                        And I tried to convert them to Cineform RAW using Cineform Studio... it works "a little".. unfortunately the Bayer pattern is different from the BMCC and so far it seems Cineform is always expecting that exact Bayer arrangement... so while import works the colors are pretty off .. but the debayering and color correction in Cineform looks great - I already wrote a support ticket to the Cineform folks in regard to that, let's see what they say. Or perhaps I have to write a conversion for Bayer rearrangement (although I can't do that in realtime while recording as it'll burn too much CPU cycles).

                        So I have to figure out now what Adobe Camera Raw doesn't like about my DNGs.. really stupid trial & error stuff.

                        And we tried some really really sharp machine vision lenses from Kowa (8.5mm 1.8) and Tamron (8mm 1.4, 16mm 1.4, 25mm 1.6, 50mm 2.8) today.. all solid metal and really really sharp image! But these things are smaaaaaaallll
                        Of course wide angle is hard with this chip size... tried a 4mm 1.4 which is about 24-26mm full frame/5D equivalent... gets already a little fisheye-e around the corners (but nothing GoPro like .. but there are a few wide angle machine vision lenses that don't distort much... we're looking for them..

                        By the way.. although wide angle is hard to do.. the upside of a smaller chip is the amazing near focus range with these lenses. You can do macro shots with basically all of them.. getting stuff 3-4cm in front of the lens in perfect focus.


                        • #13
                          Ok guys.. here it is.. some of the first cDNG that my beta cam software wrote that would even open in Adobe Camera Raw. The color science is still a little off as I haven't calibrated anything but lent the color matrix from the Decklink

                          Only 3200 x 1800 at 8 bit RAW for this... 12 bit would have been no problem at all but it was a simple user error while recording some test footage while at Tamron. Still it's not too bad I think! (with 8 bit normally even 3728 x 2096 would have been possible!)

                          Give it a try in ACR or whatever and let me know what you think!


                          I'll have more/better test footage ready in the next days!

                          EDIT.. I think this thread needs more pictures little preview.. no "real" CC or anything (try the DNGs yourself)

                          Last edited by Soeren Mueller; 11-29-2012, 05:25 PM.


                          • #14
                            Man.. Great job on what you are doing.. Really interesting to watch your progress.. Keep it going


                            • #15
                              Very interesting following this. I saw this sensor block and thought it would be a great project base, but I lack the skills needed to program such a thing. Please keep us updated!