New Cintel Scanner from BMD

mike burton Apr 7, 2014

  1. I don't know what lens the MWA scanner is using, but on our Lasergraphics, it's a Schneider-Kreuznach printing lens. Very expensive, ridiculously sharp. I'd expect the MWA is using something similar, but I'd be surprised if the Blackmagic scanner is using optics anywhere near this good, given the price of the scanner. The lens alone costs a bundle and these aren't exactly commodity parts...
     
    Robert Houllahan likes this.
  2. No an ex-collegue did but the sharpness I saw wasn't from artificial sharpening. I've worked with 2 separate scanners and the maximum sharpness seems fairly identical although I wasn't able to run the sharpness test on the second one as it was at a secure facility. The Müller scanner I used to operate had very poor optics, after we had custom optics made up the quality went up a lot. Still far from ideal, but usable. The only thing is that I don't think the optics in the Cintel scanner are top knotch and combining that with the crop imposed by the 16mm gate just isn't optimal.

    Basically the 16mm gate is just there so you can scan 16 but you need to take the scanner apart and improve the optics if you really want that 'grain' resolution. I don't have the guts (or cash) to get one and tear it apart.

    Just a side here, but is it a strange assumption that its "just" the 4K production camera with custom firmware that's doing the capturing here?
     
  3. I think the assumption is that the Cintel uses the same sensor as the 4K BMD camera.

    The 16mm windowing is pretty poor form, Bayer mask sensors really resolve about 2/3 of their stated resolution and I find that under-sampled scans look pretty bad, especially if the film stock is grainy.

    Add to the sub 2K res is the fixed pattern noise inherent in all cmos sensors, the better sensors have less but it is always a challenge with scanning negative because the hilites end up in the shadow range of the sensor where noise is most likely to show up.

    To improve the Cintel you would have to crack it open and make a system to move the camera and lens relative to the film plane, and rewrite the software so it doesn't window the sensor in 16mm.
     
    Jeff Kreines and Marc Wielage like this.
  4. hi,
    here are two scan of super 16, one from cintel, one from spirit.
    https://mab.to/Pnp91VS1U
    Cintel looks sharper, more dynamic, what do you think?
     
  5. It would depend on which Spirit and at which resolution and at which settings. There was the original Spirit Datacine, then the Spirit 4K, the Spirit 2K, and the Spirit HD -- not exactly the same. I have seen two different Spirit 4K's make different pictures due to different factory alignment procedures (especially enhancement).
     

  6. spirit hd, scan was made in a facility in belgium, I can't tell wich setting they've used. both video are provided at there original size.
     
  7. Just looking at the TIFFs I think the Spirit looks better.
     
  8. One key is that I don't believe the stock Spirit HD can do a pin-registered transfer (in software); the later Scannity can do this, using I believe the sprocket holes as reference. Just pin-registering the 16mm can have a huge effect on apparent sharpness.

    There's also the issue of color range. I've mentioned before that Imagica scanners were really problematic with colorimetry, particularly with print and IP. Not terrible with negative, but I'd give the edge to Filmlight and Arri, both of which can look fantastic when they're set up well. I'd give either the edge over Spirit.
     
  9. It looks like the Spirit 2K-4K can do stabilization, all three of the "new" Spirit machines are really basically the same, all have 4K Line arrays and mostly the same cards. On our Spirit-2K there is provisions which you can access in the DaVinci 2K for turning on the stabilization and I hove not seen issues with jumps on splices with the Spirit-2K.

    The BMD Cintel machine works on the same basic model that the Scan Station the Xena Dynamic Perf, Kinetta etc. do, that is a Bayer CFA sensor and a telecine style continuous motion transport with machine vision to look at the perfs and stabilize the image. The problems I have found is that the faster cmos sensors (typically from CMOSIS and On-Semi) have two drawbacks. The first is low S/N and a fixed pattern noise problem which typically turns up in the hilites of negative and the shadows of positive stocks. The second is that these new sensors have funny spectral response curves, I think this is due to the CFA dyes being thinner and not as accurate as the Kodak CCDs or a set of real dichroic filters. These sensors were not made specifically for scanning films and the sensor manufacturers have the job of getting sensitivity from these high speed sensors and making them workable in a wide range of applications, from industrial machine vision to traffic ticket cameras.

    I think this color dye accuracy issue is not totally new, the Toshiba 5K and 10K tri-linear CCDs in the Imagica machines have these problems and that is why the Northlight makes a better looking scan, because it has a Kodak CCD.

    What I see in the BMD scan is a sort of chunkiness and what looks like some sharpening but a lack of detail compared to the Spirit scan and that the BMD has color rendition issues. With higher res CFA Camera scanning the "chunkiness" and detail issues go away, i.e. the 4K or 5K CMOS CFA cameras can be pretty detailed but they all miss out on ultimate color fidelity and all have to hide noise to one degree or another. I would suspect that it would not be hard to make Fixed Pattern Noise show up in the Cintel scan. One of the tricks the scanner manufacturers use is to set the camera up with a pedestal to push the signal away from the worst part of the shadow response of the sensor, this works but reduces the dynamic range of the sensor and these are all barely 12-bit imagers.

    We have two sensors for our "fast" scanner, a 4K On-Semi CMOS sensor (and I may be buying one of the 5K JAI cmos sensors next week) and a 3.4K Kodak 1-Tap CCD. Guess which one makes a higher quality image? hint: the Kodak 1-Tap Color CCD does, the tradeoff is the CMOS sensor runs at up to 30FPS and the Kodak CCD runs at 5FPS. The Kodak Color CCD is better than the CMOS sensors ( and I think Jeff Kreines has discontinued using fast cmos sensors on the Kinetta due to picture quality issues, maybe he can chime in here) but even that sensor is not as color accurate as a real RGB scan IMO.

    So at the end of the day? The BMD and the other cmos fast scanners are not always suitable for work at the very high end, i.e. national and international TVCs or High end DCI scans, etc. but they can make very good pictures and quickly. Like many things in this business there is a pyramid, cost - quality - speed and the choice is as usual pick any two.
     
    Jason Myres and Marc Wielage like this.
  10. Can you be more specific? To me the cintel looks shaper, grain is thinner, especially in the sky and the skin has more details. I cant believe the scanner add sharpness and neither did I. As for color what part of the image are you refering to?
     
  11. The Cintel image is lacking detail, overall, there is detail missing in the background the sky and the foliage which is present in the Spirit scan. Them mans facial detail is smoothed over and seems sharpened to compensate for the lack of detail. In the Cintel scan the hilites are inaccurate and have a magenta tone and the mans facial tone is not natural. Also the Spirit scan resolves the film stock grain more finely.

    The BMD scan is exactly what I would expect from an under-sampled bayer mask scan.

    Do you have the original un graded Log-C scans?
     
  12. Jason Myres

    Jason Myres Moderator

    3,628
    2,267
    0
    Robert brings up a lot of good points, especially regarding grain and noise. To me the biggest difference is in the handling of the highlight values in the shirt, which aren't as well distributed in the Cintel as they are in the Spirit. If I had to grade these, getting the color and saturation of the skintone, sky, grass, etc worked out would all be fairly straight forward, but getting the Cintel shirt looking nice might require some effort.
     
    Robert Houllahan likes this.
  13. Rob writes:

    "The Kodak Color CCD is better than the CMOS sensors ( and I think Jeff Kreines has discontinued using fast cmos sensors on the Kinetta due to picture quality issues, maybe he can chime in here)"

    We only once used a CMOS sensor so far, on an experimental 4K monochrome machine. Fast but limited dynamic range. That was several years ago. Our current preferred sensor is the 4.8K Kodak/OnSemi/Truesense sensor with 7.4 micron pixels -- best dynamic range, and in our configuration we can capture at up to 12 fps full res, captured as a 16-bit RAW file. We are working on a sequential monochrome version with the monochrome version of that sensor (already designed into our hardware, just has to be implemented in software) though for most users it probably won't be necessary. Of course it will be slower. One reason for this is for a special experimental scanner for a research project with non-standard wavelengths for non-standard color film.

    The CMOSIS 20000 CMOS sensor, used by Lasergraphics and JAI, while fast, lacks the dynamic range of the Kodak CCD, and that really shows in shadow detail in print and reversal original. We've tested it, but the Kodak blows it away.

    I haven't used the BMD scanner, but am surprised they don't use their 4.6K Fairchild Scientific sensor -- oh, wait, they can't get it to work in Global Shutter mode, so I am not surprised -- because it has significantly better dynamic range. Too bad.

    BMD themselves is telling potential buyers that it's not designed for archival film -- it's intended for the shrinking market of new color neg. And, really, 35mm color neg, because scanning S16 at slightly less than HD res is insane!

    The cheap Lasergraphics does the same thing -- fixed sensor and crop for the format -- scanning 8mm at 1K is a travesty! See this for why: 5SlAVkbT

    There are some interesting CMOS sensors coming soon from Sony -- the Pregius line -- which have better dynamic range than other CMOS sensors, and are fast -- we may use one of these in a new lower-price scanner later this year.

    Note that the sensor module in the Kinetta can be swapped out in one minute, so as newer and better sensors become available, you aren't stuck with technology from 2 years ago that you can't easily upgrade.
     
    Robert Houllahan likes this.
  14. We are using the 4.8K Kodak in Monochrome (Oxberry Pin Registered) on one Xena machine, it is a fantastic sensor with exceptional dynamic range and extremely low noise. We recently did a 35mm 8-Perf Vista Vision develop and scan on that machine and the results are top notch. In monochrome we get about 2sec/frame in 1-Tap 14-bit true RGB.
     
  15. Jason Myres

    Jason Myres Moderator

    3,628
    2,267
    0
    For fun and to see how far apart these were, I did a basic grade, and created screen grabs of their settings. I used a single primary with a little HueSat to help match the skintones. I've included TIFF frames for each so you can compare them on your screen.

    https://s3.amazonaws.com/liftgammagain/attachments/KUBE_Cintel-Spirit_BasicGrade.zip

    [Update: Looked at these again when I got home, and found a better (brighter) sample frame that allowed for a more minimal grade with a lot less adjustment.]

    The main adjustments for each were:
    - Offset via Printer Points (Left Hue Wheel)
    - Brightness
    - Contrast
    - Shadow/ Midtone / High Levels
    - Highlight Hue (Right Hue Wheel)
    - Hue Vs Sat (Curve)

    What each was used for:

    Offset: match overall color cast
    Brightness: initial overall level adjustment
    Contrast: used in conjunction with Shad/Mid/High to help match levels
    Sh/Mid/Hi: after contrast, used to help smooth levels with a focus on the shirt and sky to soften clipping and bring out detail
    Highlight Hue: used as a fine adjustment on the highlight color cast
    Hue Vs Sat: used to match skintone saturation

    The shirt did take the most time. It contains the brightest highlights in the shot, and needed protection from clipping without flattening the details in the fabric.

    Overall they are really quite similar, but the Spirit seems to have nicer balance in terms of levels. I would love to see the greyscale curve on the two scanners since the Spirit was very log-like, while the Cintel has a lot more initial contrast, but less saturation. Like Robert mentioned, the grain in the Spirit is also a lot more evident. By comparison, the Cintel almost appears as if it's been de-grained.

    There also seems to be stretch in the aspect of the Cintel frame. I could be wrong, but the Spirit aspect looks correct to me. If you A-B the TIFFs in the link above it's easy to see. Maybe someone can identify what's happening there.

    Cintel Primary
    Cintel_1.jpeg

    Cintel Hue vs Sat
    Cintel_2.jpeg


    Spirit Primary
    Spirit_1.jpeg

    Spirit Hue Vs Sat
    Spirit_2.jpeg
     
    Robert Houllahan likes this.
  16. Great Jason.

    Just a quick observation, this film image is a fairly easy target to hit, I presume this is a low speed daylight stock which is well exposed. I have found that scans which are under-sampled really fall apart quickly with more grainy stocks and that is especially true with bayer mask color scans when compared to the full res 1080p or 2K scans where each color layer is sampled at full resolution. I think this is why this scan has a lack of detail, the undersample sort of rounds off the detail but as the detail becomes more fine the image becomes more chunky.
     
    Marc Wielage likes this.
  17. hi guys,
    three CRI ungraded of three different scan of that same image:
    https://mab.to/cFs1LLthV

    have fun :)

    PS:
    S16 once cropped of the cintel is about 1920x1152 wich is a bit more than 2k DCI 1.66(1792x1080)
     
  18. Just remember that S16 scan is 960 pixels of Green and only 480 pixels of R and B I think it is a fairly well understood norm that a Bayer mask CFA camera needs to be over-scanned about 30% in order to resolve the target resolution and 50% is better.

    See Mr. Nyquist:

    https://en.wikipedia.org/wiki/Nyquist_frequency
     
    Al Arnold likes this.

  19. Could you share the CRI files again?
     
  20. hm, can't find them again, sorry...
     

Share This Page