Personal View site logo
2K BlackMagic Pocket Cinema Camera, active m43, $995
  • 4493 Replies sorted by
  • For once I stopped looking for white orbs and just enjoyed the little film :)

  • @vicharris Well I tried, I really tried, but now I just see bad highlight roll-off in general. That being said I think this is going to turn out to be less a sensor issue and more a calibration or interpretation problem within the software. Its as if the exposure log scale ends abruptly in the highlights and immediately clips all the detail. I honestly don't see how a hardware issue could cause this unless it's due somehow to internal reflections that concentrate around the highlight areas on the sensor and amplify the respective light intensity.

  • @Tron, the way the Fuji issue was explained, it's the sensor photosites overloading (fine), but then spreading the overload to neighbouring photosites (not good). That's why it's always pretty much perfectly circular, at least when a single dot highlight is involved. The more complex shapes also seem to be made up of several circles overlapping.

    I don't know if this is just a sensor issue, or some interplay between sensor and optics (eg. the covering glass that was mentioned), but you'd think an optical issue would be a more blurry bloom (optics usually introduce some kind of smear right).

    I also think it's unlikely to be a software clipping issue imo. If you consider that Fuji orbs example:

    On the good sensor there is a gradual falloff and a lens flare - if you clip that, you would not get a perfect clipped circle, because the flare is non circular. I've attached a simulation (clipped in Photoshop).

    fake clipping.jpg
    520 x 780 - 150K
  • I just liked the video :)

  • Actually I'm no longer sure. My simulation isn't right, because the original scenario is a high dynamic-range environment, it can't be simulated by clipping an 8bit image.

    I guess it depends how bright the lens flare is vs the original super-bright light point at the center. If the point is much brighter (probably), then I guess it might be clipped in a circular way.

    Hmm... any new word from BM?

  • Created a short comparison between the GH2 and Blackmagic Pocket Cinema Camera here:

    Here's my explanation/review that I also posted on the vimeo page:

    Conclusions

    1) Although the German video site Slashcam found the non-artificial sharpness in the Pocket to be very high, it seems to me that the GH2 still resolves better image detail. (Not entirely surprising with an image downsampled from a 16 MP sensor as opposed to a debayered 2 MP image.) However, the GH2 video image would have been inferior resolution-wise if both cameras had been tested at 800 ISO.

    2) In terms of dynamic range, gradation and shadow detail, the Pocket is in a different league than the GH2. This is already visible in the video image that has been graded to look like the GH2's. The possibilities to grade the image to match other looks/cameras are very limited for the GH2 and comparatively endless with the 10bit native ProRes material from the Pocket.

    Caveats

    ...for owners of the GH2 and similar (amateur) cameras: operating the Pocket is much more demanding in every regard: exposure, focusing, white balance.

    1) With its high native ISO, the camera absolutely requires the use of ND filters. They also are the best recipe against the camera's white orb/blooming problem.

    2) The focusing aid zoom (for manual focusing) is less precise by design because it only zooms into a 2MP instead of a 16MP sensor; and while the camera offer focus peaking, I principally don't find peaking reliable enough for critical focusing.

    3) Exposure aids are limited to zebras and, for electronic MFT lenses, an auto-aperture button. Thus, the camera encourages exposing to the right, which is not a good strategy if your scene has a high range of brightness (say: a street at night with cars driving by) and you need to let some highlights blow out. Highlight roll-off is harsh, and provokes white blobs if you overexpose by more than 2-3 stops. Conversely, there is no histogram or zebra to indicate underexposed parts of the image.

    4) White balance is adjustable in a few preset Kelvin values, which is fine for photographers/cinematographers used to Kelvin temperatures for different film stocks. There is no possibility to set white balance by locking the camera to a neutral grey object, and no possibility to compensate for green/violet tint (a limitation when recording ProRes).

    5) If you want a good image, don't use Panasonic's and Olympus electronic Micro Four Thirds lenses, at least no zooms and no focal lengths below 45mm, because their optics distort, have purple fringing and require in-camera software correction. That also rules out all available stabilized MFT lenses. If you want to shoot handheld or with a simple handheld stabilizer, your focal length should be 14mm or less, and 17mm maximum. I found the SLR Magic 12mm/f1.6 to be a good all-round lens for the camera.

    Older 16mm cine lenses, such as the Canon TV16 13mm/f1.5 or the Kern Yvar 14mm/f2.8, work as well, but don't resolve as good as a modern lens - and they are among the very few wider angle 16mm cine lenses that cover the S16 sensor of the Pocket.

  • Nice observation @cantsin

    Two things...I think if you were to add sharpening in post to the bmpc clip to match GH-2 you will see more of a match or maybe even more resolve in the the pocket camera. This camera can use the post sharpening. But the second thing, which is also evident with the BMCC, is that ProRes is not nearly as sharp as RAW. So I can safely bet that the images would turn out a great deal more detailed if processed in RAW. (jm2c).

  • Yep, tons of room there to sharpen in post. Nothing surprising.

  • I think if you were to add sharpening in post to the bmpc clip to match GH-2 you will see more of a match or maybe even more resolve in the the pocket camera...

    This shouldn't really be possible due to the Nyquist Theorem. The only thing that should really allow for this would be if the GH2 were running a patch of sufficiently low quality to allow the compression to smudge away its luminance resolution advantage in the spatial domain.

    It's like one of the problems you see in digital effects and motion graphics added to scanned film and digital video. Most amateurs and lower-grade professionals forget that you invariably have to add a blur filter to rendered CGI layers when adding them to existing plate photography, if they're rendered at or above the resolution of the photography. This is due to the fact that even moderate CG oversampling tends to default to 4X.

  • Not sure man. I'll have to say I can add much more sharpening with my BMCC footage than I would ever dare to with GH2 stuff. With GH2 footage, it loos bad really fast but just playing around with hair and sharpening with BMCC, it still looks pretty damn good. It's so easy to bring out the eyes with this footage and preserve everything else. Doing it right now with my latest promo I shot.

  • @Ian_T - the Pocket footage was sharpened in post (with an unsharp mask); I also mentioned that in the technical specs on the Vimeo page for the video.

  • @vicharris the trick is you don't use a traditional sharpening filter. Those are garbage. You also don't apply sharpening to the image as a whole, both chroma and luma. But 10bit 422 isn't a panacea to allowing gobs of extra sharpening to be applied since 444 isn't either.

    This was lamented to me by David Slade who was quite disappointed with how soft he thought his Alexa ProRes stuff from Hannibal looked. I thought he was smokin' crack because his episodes look fabulous and most folks would be seeing highly compromised, compressed and/or scaled episodes anyhow. So he really pushed as much as he could through on his episodes using a package designed for stills photography (NIK Sharpener) since the typical sharpening filters for editors and most post software are junk if you're not willing to "roll your own."

  • @BurnetRhoads Hmm, good info. Thanks. I agree with the basic stuff. While it's by no means pro level but it is a step above basic sharpening, I usually use NeatVideos controls to give me a little more control. Once again, I know it's still low level stuff but I fell I can pull out eye detail alone with it and not really touch anything else. Could be me just convincing myself I'm doing something different though :)

    Just looked up that sharpening program. Pretty cheap if it works like that.

  • Also, with GH2 you're already starting out with a sharper image so it needs less post sharpening. Basically I don't work to bring out any structure or edges, it's tiny bright details and tiny black details that make the image pop when you apply local contrast enhancement techniques carefully, to just the luminance channel.

    My brother is currently editing a proof-of-concept we shot where I was on A-Cam and a friend was on B-Cam with his Canon DSLR and he was shocked, even after we did a feature on the 7D, how soft the Canon stuff is. He and the other producer have listened to me talk schmack about how much better and sharper the GH2 is but this is the first time they've ever seen it side-by-side. It's got to be pushed to try to match the GH2 Moon Trial 3 footage and then I'll have to be careful that it takes even the mild enhancement I'd normally do the the GH2 footage, since it doesn't really need overall structure sharpening at all.

    edit: oh, and I've never used Nik so I have no idea how easy or hard it is to work into someone's pipeline compared to Neat Video which is no biggie. Personally I use a manually constructed stack of effects in AE that does something like Unsharp Mask, but it doesn't false-edge in the same way that filter wants to do, feels faster, etc. I came upon the technique while researching what went into the "Dragan" look popular in commercial photography. Mostly I was interested in what made all the tiny bright and, especially, tiny black details really pop. Skin pores and stubble and that sort of thing. It also really brings out highlights in eyes.

  • @cantsin

    I find it hard to believe that the BMPCC footage can't be made to look as sharp as the GH2 footage. The question here isn't of absolute resolution, since we already know the BMPCC provides substantially more measurable resolution, thanks to nomad's charts.

    Could you provide an uncorrected/unsharpened high resolution exported still of the BM footage? If so, thanks!

  • I thought Olympus lens have less purple fringe because their camera don't do CA unlike Panasonic?

  • @jrd if the sensor is purely 1920x1080 it is impossible for the pocket camera to equal a GH2's true image detail, all things being equal. They would have to be doing some kind of pixel-shift techniques to overcome its lack of super-sampling, which shows up as aliasing and moire.

  • @BurnetRhoades

    But the GH2 doesn't resolve the full 1920x1080, any more than Canon DSRLs do. Look at nomad's charts earlier in the thread.

  • @jrd that's a set of charts and one example with conditions the settings of which haven't been confirmed, fully disclosed or recreated, that I'm aware. Observers also haven't taken into account the difference in performance between filtered and unfiltered imagery put through additional scaling and re-encoding. Even Nomad himself expressed doubts about the conclusion, what I read, and agreed that some of what was seen was false detail.

    Results have to be reproduced and all of the factors leading to the conclusion explored before something is a fact. The only fact here is that known signal processing theory precludes the Pocket Camera from resolving more than the GH2 without some mitigating factor destroying the GH2's basic advantage of being supersampled. The rest is a lot of assumptions and desire for an easy one-line answer.

    Then you have to account for processing that's going to be necessary on the PC to remove aliasing and moire, that won't have to be performed on GH2 footage and take that all the way to the end. Most of the time you won't need it though, I'm betting, at least not the antialiasing. For now the only conclusion I'm willing to accept is that the extra clean nature of the non-compressed footage carries a higher perceived level of sharpness in some instances while at the same time provoking comments like "soft" in numerous postings of footage since its release. Something that isn't common at all for the GH2, hacked or otherwise.

    edit: and I'm not dogging the PC. I think it's very cool and will come through Blobgate okay. If my GH2 were to break or I'd otherwise need to replace it I would certainly do so with a PC versus a G6 or GH3 or any other DSLR, if I didn't just get a full-size BMCC as replacement.

  • Has there been any footage that's been shot with the PannyLeica 25mm? It's my favorite lens (that I own) on my GH2 and I'd be disappointed if it turns out like crap on the BMPCC... I also have the Panny 14mm but I don't really expect much out of that lens anyway.

  • @BurnetRhoades

    Based on experiments with J. Brawley's Prores footage, I'm persuaded I can make anything shot with the BMPCC look better than the equivalent on a GH2, including equal or greater sharpening, if that's the goal.

    The only way to find out is to try -- maybe cantsin can provide a few of the original ProRes frames.

  • The BMPCC definitely has aliasing and that means false detail. Sharpening will only make that worse.

    The GH2 is not free of aliasing, but it's mainly in the luma channel. Hardly any chroma aliasing, which is massive in the BMPCC.

    Making the stuff look better than a GH2 means using the better gradation of subtle color transitions when grading, not sharpening the hell out of it.

    BTW, my tests can be reproduced by anybody with both cameras at hand and a standard ISO 12233 test chart, which you can download as a vector file at Cornell university for free.

  • @jrd here are TIFF screengrabs from the gh2 and the pocket test footage, both created with ffmpeg directly from the camera source files. gh2.tif has 8 bit color depth, pocket.tif 16 bit (logically):

    gh2, 5.8 MB

    pocket, 9.6 MB

  • @MagicMountainMan Be prepared for disappointment, because the PanaLeica 25mm has native barrel distortion that is firmware/software corrected in cameras like the GH2 but not in the Pocket (since it can't pixel-bend a 2MP image without degrading resolution). Besides, 25mm is really too long a focal length on the pocket for anything but tripod use. For handheld/rig-stabilized shooting, 12-14mm is ideal.

  • @cantsin What's wrong with barrel distortion? I think it's partly the lack if it in modern lenses that makes stuff look rather anodyne and what makes stuff shot with older/manual lenses look way better. Also since the sensor really only sees the middle of the lens I am imagining it won't be as bad as the edge in terms of CA. But that's merely conjecture. Do you have examples of stuff shot with Panny lenses that show the quality drop. I'd be interested to see since I own quite a few Panny lenses and have a Pocket on order