Personal View site logo
How to build Facebook's open source 360 Surround camera
  • 12 Replies sorted by
  • Any thoughts on this camera system? I've built two 360 rigs - one with 6 GoPros and one with 4 GH4s (really fantastic, but not stereoscopic). This one seems to have some problems -- lots of cables, maybe too many cameras to stitch, not really stereoscopic at top and bottom.

  • I think the best option by far is 6x Xiaomi Yi cameras. Next step up? Throw a few more Xiaomi Yi cameras at your set up!

    My (semi realistic) dream set up that I'd like to step up to is 6x BMD Micro Cinema Cameras. As of course BMD cameras would provide a huge leap forward in image quality over Xiaomi Yi cameras, while still keeping to a relatively small form factor.

    My unrealistic dream set up would be: 6x KineFinity Terra 5K (which you could build for only very slightly more than the cost of Facebook's one in that video!!).

    I've followed closely the many other options, and none of them really interest me, not the Nokia Ozo, not this, none of them. With a few exceptions: Samsung Gear 360, have used this a number of times, it is a bit flawed and not the best specced, but if you need a very small (waaay smaller than Xiaomi Yi / GoPro one) set up (such as on a small drone) then this does it well and at an attractive price. Nikon KeyMission, it isn't out yet, but interested in it for the same reason as the Samsung Gear 360. (and hopefully with be better than the Samsung Gear 360?? Wishful hoping) Yi's Google Jump, the low price of Yi cameras combined with the power of Google Jump? Has my interest!

  • @DouglasHorn, remember you also need a custom PC with 5!! USB3 expansion boards for all those USB3 cameras.

    I would really like to see a few comparisons between this and other 360º capture solutions. The other problem is that many of these camera systems use Machine Vision cameras that are not optimised for cinematic video capture.

    @DouglasHorn really interested in your GH4 360 Rig and the lenses you went with! Please share if you can.

  • I think I saw somebody share pics of a G7 360 rig somewhere? That would also be interesting.

  • So a major difference between systems is whether you want to live-stitch/stream your footage or not. If you are not live-stitching, then you can just load footage sequentially and stitch in AVP. For live stitching, there are a couple of options, but the one I'm most excited about is the new Teradek Sphere with 4 HDMI inputs that can be chained to 8 inputs with a second unit. You stitch and stream via a iPad Pro. There's another solution available as well.

    My GH4 setup is evolving. Currently it consists of 4 GH4s in a tetrahedral mount that I designed in CAD and 3D printed. (What an amazing time to be alive, btw!) There are a few metal parts for strength, etc. I'm currently on rev 12.2 of the mount design and figure that I'm probably 1-2 major revisions away from what I want to finalize (and probably always will be...the nature of development) But what I have now is usable and I've run it on a few different shoots. My goal now is to maximize usability, further collapse the size of the rig, and allow better cabling paths for future live streaming or other options. This is a design in-progress and I hope to employ it in some business applications, so I don't want to spill the beans quite yet on the lens/capture mode/setup recipe that I'm developing, both because it's early and may be misleading at this point, but also because I've invested a lot of time and testing in this (I've tested at least 6 different lens options for this among other things) and am not quite ready to open-source it. Though I may. I definitely appreciate the help I've received from the PV community and want to give back when I have something worthwhile.

    The best thing about the GH4 based solution is that I can set common exposure, aperture, etc. values and they stay. The quality of the footage is head and shoulders above GoPros. Additionally, GoPros just aren't rock solid in operation -- the 6-camera rig I use constantly has problems where one camera won't fire, thus destroying the take for all. GH4s are much better. I am working on timecode synching them using the app or a YAGH and Zoom 8 recorder for timecode. Again, it's all a work-in-progress that I have to dedicate time to along with my other work, so it progresses slower than I'd like.

    I will try to share some pics or 360 videos soon.

  • I'm a loooong way from upgrading my Xiaomi Yi rig, but when I do I am thinking to go with BMMCC. Because I'd prefer the dynamic range, bit depth, and colorspace of the BMMCC. And while the GH series does a max of 30p @ 4K, you are gaining resolution but giving up frame rate and rather keep it at 60fps. Presumably the GH5 will give 4K 60fps, but that will probably cost $2K or more per camera.

    Though given how long until I'll be upgrading, we'll surely see a Panasonic G7 sale by then which might make that option too cheap to resist vs going with BMMCC.

    Anyway, are you willing to share your thoughts about lens options?

    My thoughts have been the options are:'

    a) native mount (i.e. MFT lens, or E mount lenses if a person was using A6300/A7 series instead), but then this can severely limit a person a couple of years down the track if they change bodies. It would be a lot smoother / cheaper if only the bodies need to be swapped out and not lenses as well. I'd like to have some degree of versatility with this rig.

    b) full frame UWA lenses, not an option as you can't then go truly wide if using them on APS-C or smaller bodies. And A7r II / A7s are the only truly interesting options to consider here for use that are full frame. Maybe with the one exception of the Rokinon 14mm f2.8 lens, which does almost hit the sweet spot for price & focal length even when used on APS-C. Or the Rokinon 12mm F2.8, but then you have to deal with fisheye distortion.

    c) APS-C UWA lens, this appears to hit the sweet spot of maximum versatility plus maximum FoV.

    APS-C lens options:

    Sigma 8-16mm f/4.5-5.6, max FoV but I'd be concerned about the slow f-stop, as then you'd start to be losing one of the key gains of ditching GoPros with their poor lowlight ability.

    Tokina 11-16mm f/2.8 (or the newer Tokina 11-20mm f/2.8, but that costs more so no), this is the lens I own myself and is in my eyes the "best" UWA lens for normal filmmaking, but does that mean it is for 360VR too?? Hmm

    Rokinon 10mm f/2.8, is lighter/cheaper/wider than the Tokina but is a fixed focal length (probably not a disadvantage though at all!). However isn't cheaper than Tokina when you consider the older Tokina models can easily be picked up secondhand, but the Rokinon can't be so easily found at all secondhand.

    Rokinon 16mm f/2, the fastest option but by this point at 16mm it is only barely UWA at all.

    Outside these options listed, I can't think of any good UWA lenses, or am I missing something? They seem they'd all be worse compromise somewhere in price/FoV/speed/etc than these.

    What are you views on the thought process and each of the options I reached at?

  • While doing some googling I stumbled across these guys also using GH4 cameras for 360VR: http://shinichi-works.info/project_gh4.html

  • @IronFilm - I had seen that video at one point in my research. They're using the Panasonic 8mm F/3.5 fisheye. I'm pretty sure they are not getting full 360 degree coverage top and bottom. (Also, that design is really flimsy and wobbly which is a killer for 360° video stitching.)

    I tried several of the lenses you mention and ultimately discounted them. I believe that fewer cameras are actually better provided that you have good overlap. That meant that I narrowed my search to true circular fisheyes with ≥180° coverage, which is what I would recommend to anyone. There's still a distorted coma around the edges that's unusable so you get less overlap than you'd like. I tried a version of the rig with just three GH4s, but although I was technically getting 360° vertical coverage, there was not enough there to stitch well (you need overlap for the software, but also the imperfect edge of the lens ate into the image). The tetrahedral design is really key. Side note, I am adapting my GoPro rig (using 220° lenses) to a tetrahedral design as well. It's harder to build the rig, but overall, much better.

    Another thing I've learned is that the amount of visual data we are capturing shooting 4-6 cameras at 4K is probably overkill. The problem is that by the time these videos actually get in front of your eyes, the pixels are really squashed together, just due to the physics of current displays and ocular distance. I think it's likely to be that way for a while. I'm experimenting with capturing each camera in 1080P instead. That will enable higher frame rates.

    I hope this helps narrow your lens search a bit. If you ever get to the point where you're really going to build one and want more info, go ahead and send me a PM. I may have more information at that point.

  • @DouglasHorn

    4K small screens, and most probably even 6K, will be standard feature very soon.
    For safety you need around 15000-16000 horizontally for 360 degrees.

  • So Metabones XL (or regular RJ Lens Turbo is enough?) with Rokinon 8mm fisheye is the way to go?

    Interesting you're such a fan of using a minimal number of cameras, as I'd have assumed the more cameras the better!

  • @Vitaliy_Kiselev - True, and I currently output everything at the ~4K 360° spec just for future proofing. (Not that anything I'm filming at the moment is really worthy of future proofing...) However, my real point is that shooting 4-6 cameras at 4K resolution then extracting ~60% of each image into the stitching area means that there are a lot of extra pixels to crunch that do not currently make much difference in display. Remember that you are essentially cutting the display in half and serving an even smaller portion of that half to each eye. On my Samsung Gear VR viewed on an S7, the pixels still look incredibly large--like a color TV from the 1980s. Displays will have to be a lot better before this problem is solved, due in part to the lenses and very short distance to the display.

    For current stuff, I think that capturing in 1080P is probably more than enough. ...but for now I still capture in 4K anyway.

  • @IronFilm - Yes, to me fewer camera is actually better since you aren't stuck with a huge data glut. Also, each new camera introduces the opportunity for something new to go wrong.

    Perhaps I'm very wrong for some reason, because all the big pro camera setups seem to emphasize more cameras. I did find that my 6-cam GoPro (220° lenses) solution was better than 4-cams for the same flat configuration, but I expect that 4 camera in a tetrahedral array will actually be better than either. Again, it's just a harder rig to CAD and build. More cameras are necessary if you are using stock GoPro lenses, of course. More cameras do help if you want stereoscopic 3D in addition to 360° video. But the solutions I see don't work above or below the horizon except for the Nizo.

    I definitely explored the Rokinon 8mm fisheye. I did not use that in my rig with or without a focal reducer. It is a rectilinear fisheye, and I believe you really need a circular fisheye. But you may have better luck with it. My processes involved looking at the various modes currently available on the GH4 for capturing to the sensor, then pairing those with the circular fisheye lenses of 180° or more that I could mount to the camera (fewer than 10 lens options). My options came down to 2-3 finalists that I could get to project the entire circle onto the sensor and I tested and chose between those based on price, quality, ease of use. In the end there was one very clear winner. I feel like with what I've said here, any semi-regular PV reader could decode my methods in about an hour of looking at lens specs online -- without me necessarily just giving away the answer to everyone who knows how to type in Google.