My Gopro 3+ Black 180 degree stereo rig

A new megapixel ultra fisheye from CCOM -- $99 -- 220 degrees

http://www.megapixellenses.com/sell-244 ... -lens.html

http://www.youtube.com/watch?v=W51gGi0bGEo (NSFW language)
http://www.dropbox.com/s/lr2l7evru0f44j ... 80_3dh.mp4 (will work directly with GearVR if you put it in your MilkVR folder)

This is a bit of immersive journalism (180/180 equirectangular 3d) -- stabilized with PTGui. PTGui is good for this because there is distant background visible throughout the shot and the camera is not moving around much in space so the background is effectively infinitely distant. I write a bit about stabilization with PTGui and when its useful in this thread here:
viewtopic.php?f=33&t=18491

... in that post where I write about memory problems with stabilising PTGui sequences longer than a few hundred frames ... PTGui can use graphics card memory and it can increase usable frame sequence length a lot:

Options/Enable GPU acceleration: If enabled, PTGui will attempt to do image processing on the graphics card (GPU). A powerful GPU can speed up stitching several times. To use the GPU for stitching a video card supporting OpenCL 1.1 is required, with a minimum of 512 MB video RAM. At least 2 GB of video RAM is recommended: although the panorama doesn't need to fit entirely in the video RAM, moving data between the CPU and the GPU is relatively slow and should be avoided to get the best performance. Every recent GPU from AMD and NVIDIA is supported. Integrated graphics (a graphics processor integrated in the CPU) are not supported; in general these would be slower than stitching on the CPU anyway. On a Mac PTGui requires OS X 10.9 (Mavericks) or later for GPU stitching.

This is processed btw using the Dehaze filter new to some Adobe image programs -- like Adobe Raw. It is great for improving the detail of Gopro Protune video (after converting to frame sequences) without adding grain and artefacts.
https://www.youtube.com/watch?v=bmkJpCg_BJ4

This is shot with a 5cm interaxial and a lot of it is extremely close to the camera and there are stereo errors in some areas -- like in the Roman centurion's elbow on the left. I have since developed a new 3.5cm interaxial rig for this kind of close quarters reportage situation.

What software are you using to warp your footage to equirectangular? Ive been doing tests with non modified gopros and haven't found a way to warp them into a way I can look at it in 180 dome.

What software are you using to warp your footage to equirectangular? Ive been doing tests with non modified gopros and haven't found a way to warp them into a way I can look at it in 180 dome.

With PTGui -- you would need to convert the videos to frame sequences first -- Virtualdub etc -- and then load the frames into PTGui (there is a demo version). You need to set the Image type to full frame fisheye, and the fov to 125 degrees. Then set the panorama output type to circular fisheye fov 180 and set the pitch in the image settings to 90 degrees and in the output set to individual layers.

Re ultra fisheye mini lenses -- folks might recall my mention of this upcoming M12 220 degree lens
http://www.aliexpress.com/store/product ... 26520.html

.. you should note if thinking about it that they have added now info that it DOESNT fit Gopros -- presumably because it cant get close enough to the sensor with its shortish lens barrel. Still an interesting lens if it really is 16Mp. Maybe good for Mobius for instance as it would probably give at least 180+ horizontally on that format and it is easy to hack the focusing mount of that camera.

Re Mobius I notice they have a new model the Mobius 90
https://www.mobius-actioncam.com/store/ ... m-package/
This is the same as before except the sensor is rotated 90 degrees so you have to have the camera vertical (like on the side of a helmet) to get normal letterbox video. The Mobius is a very skinny camera in this orientation giving you the possibility of an extreme dense circular array.

The Theia 16Mp 180 lens is now on Bhvideo

http://www.bhphotovideo.com/c/product/1 ... on_ir.html

I am keen to hear reports on this.

More news:
The Xiaomi genlock thread has recent contributions:
https://forum.dashcamtalk.com/threads/x ... ras.12175/

Here is a recent Facebook stereo panoramas group post of mine on stereo still panorama capture:

Here is a video (up to 4K resolutions available) which illustrates some of the logistics of capturing stereo still 360s with a 3 camera modified Gopro rig.
https://www.youtube.com/watch?v=JCyrTnmo0RU

with my 90mm interaxial rig .. this one

https://www.youtube.com/watch?v=Yzix8a_PqCg


The first two videos are L and R (90mm interaxial) and the third one (in the row) is the central camera used for calibration. It is shot from a Nodal Ninja pole at 6m. Sometimes I use this pole with my twin 5DMkII fisheye rig (shooting synced stills) but I need a bigger rotator for them with more nadir obstruction and it makes me very nervous to use at 6m if there is any sort of wind. The Gopro rig catches the wind less and can be spun faster with still plenty of frames and it would be less lethal if the camera rig fell on me (or someone else). I use a very heavy duty custom tripod/levelling head/clamp system to hold the pole.
The L and R cameras are synced by the Dual Hero System. The central camera is synced with the others by a sound spike (hitting the pole with a key) and aligned with the L/R sound tracks in After Effects. Here it is almost half a frame out of sync (20ms) with the L/R pair but for calibration purposes with a smoothly rotating rig it doesnt matter.
Here the rotation is about 8 seconds, meaning 200 frames for 360 (25 fps). You will notice that in this bright light there is no discernable motion blur despite the fast rotation. It is shot with the 2.7K Medium video setting on Gopro 3+ Blacks, using Protune with -0.5 EV setting. The lenses are 1.29mm 195 degree CCOM Technology M12 lenses. The videos are not edited color/contrast-wise for this exercise. The final pano resolution is about 5000*2500 pixels (ie. 10000*2500 SBS).
The first few rotations have the view masked vertically a lot to exclude the sun, the last lot, have just a bit of vertical masking at the zenith (as I know it will be easy to use content-aware fill etc to fill the hole and the zenith masking might add a bit to reducing flare).
You can see that the high viewpoint and the multiple rotations and the fairly even light gives plenty of scope for compositing multiple frames of people to make the scene more populated.
Now I will have to start stitching it to show you the end product :-)

Facebook is a dubious platform for technical forums I think but there is quite a bit of info on this one -- you need to apply to join but it happens quickly I think .. there are posts on 360 video and still stereo panorama techniques

https://www.facebook.com/groups/3dstere ... lhc.recent

Another post of mine from the Facebook 3d stereo panoramas group: about this video:
http://www.youtube.com/watch?v=TiQIKGb_vSc

https://www.dropbox.com/s/o6by6muf2atxp ... 80_3dv.mp4

https://www.facebook.com/groups/3dstereopanoramas/

I have doing some more experiments with 3 camera Iximage-style "omnipolar" hemispheric stereo video capture.
http://imgur.com/uAYyb3x
This rig has an effective interaxial of about 5.5cm. I have other similar rigs with interaxials of about 3.5cm and 9cm. Two of the cameras (Hero3+Black) are synced with the Dual Hero System, and the third is synced with sound spikes in After Effects. The lenses are 190 degree CCOM Technology lenses. I usually shoot at 25 or 50 fps at 1440p Wide with this rig.
You can capture quite good hemispherical 3d videos (straight up or down) videos with this approach and if you use ultra fisheyes you could get up to 280 degrees vertical coverage theoretically (with the Entaniya lenses, for example). Iximage have another Gopro three camera ultra fisheye rig on their home page:
http://www.iximage.com/
As compared with most other approaches this gives good horizontal stereo hemispherical stitching relatively easily with just three cameras, but with significant vertical parallax issues with many subjects. Iximage have a technical paper here:
http://www.iro.umontreal.ca/~chapdelv/…/couture-iccp2013.pdf
and a graphic pdf about stitching here:
http://nasv3d.iro.umontreal.ca/chapdelv/…/360-s3d-camera.pdf
They have their own stitching software but PTGui masking tools can do the equivalent -- with green ("keep this") bow tie shape masks doing the heavy lifting and some red masking to deal with the lenses visible in the other camera views.
There are some very particular situations where I think this approach might have great utility which is why I keep coming back to it. For instance consider a downward hemisphere capture -- all the cameras are looking at much the same scene so exposures are generally very close for each camera. And the scene is high contrast with no sun etc in the scene.
I think it could be very good for aerial stereo panoramic videos with high resolution fisheye cameras -- or giant crowds from high poles with a single mono shot for the sky. I will post links to some recent efforts shortly.
Here is one frame 2048*1024 anaglyph equi from a movie with the rig:
http://imgur.com/C7CGaNU

I have talked about this before -- correcting vertical parallax errors with side by side 180 3d rigs -- but now I have recast it as a thought experiment:

Re alignment generally for 3d 180 side by side capture: suppose you have two cameras, parallel, 180 lenses. And you have calibrated them on pano heads and side by side so you have physically accurate calibration and distant scenery will be identical in calibrated equi form say, in both cameras -- across the entire field of view of both cameras.

Now for this thought experiment imagine increasing the interaxial to say 50cm. Still distant scenery is still identical in both calibrated equis.

Now imagine say a box of cigarettes at camera height 45 degrees to the right and in front of the rig. Say 2 metres away. Obviously it will be bigger in the frame vertically in the right image equi, smaller in the left image equi. And conversely with a cigarette packet on the left side of the rig. These scale vertical parallax errors for oblique close areas are unavoidable with significant interaxials.

Now consider the cigarette packet on the right 2 meters down from its camera level position. Now the right camera is looking down on that packet more than the left camera. So this results in vertical parallax for the close to the camera oblique downward view areas. And you can see there is no perfect way to correct this without disturbing the perfect alignment of distant background areas.

And conversely for the left hand side with downward oblique areas there -- the vertical parallax errors are in the other direction.

So this is the difference between rotation slice assembly stereo panorama capture and static parallel 180 3d rig capture. The former can reduce sometimes these oblique vertical parallax and scale errors similarly to how it minimizes horizontal stitching errors.

In fact though it is possible with many side by side 180 3d videos to correct these oblique error areas by warping vertically one or both equis locally -- with the Puppet Warp tool in After Effects for instance. This enables you to fix areas which to dont want to shift at all and just warp left and right of frames for close up details where necessary. (It is in mapping terminology a "rubber sheeting" application eg. where you want to align a map to an aerial photograph.) And that way you can often much improve the stereo viewability to the sides and oblique corner viewability.

eg. this market scene:
http://www.dropbox.com/.../sydneymarkets180x180_3dh.mp4

Doesn't this also affect the background along that line of sight? In other words, don't you have to apply the stretching ahead of time to everything according to optical axis offset?

Doesn't this also affect the background along that line of sight? In other words, don't you have to apply the stretching ahead of time to everything according to optical axis offset?

if I understand you ... maybe not if the backgrounds of the problematic vertical parallax areas are textureless eg. the white walls of a room