3d generated imagery with depth map

I'm going to look into some manual approaches - taking renderings and doing post work to get them working

Many renderers will output equirectangular renders of scenes. Unfortunately the Cycles renderer for Blender only supports full 360 by 180 renders -- whereas for stereo panorama assembly a 1 degree by 180 render might be very useful.
http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Cycles/Camera

Assembly can be done with Stereophoto Maker which has a slice assembly feature for making stereo panoramas ... File/MosaicImages
http://stereo.jpn.org/eng/stphmkr/

Mediavr! Another super helpful post - although looking at that software for stitching slices - it seems like even a program like Photoshop could work in that situation. Photoshop might work even better for myself because I'm unsure how to create prefect 1 degree slices in my program, so I'll render slices with a bit of overlap and try to stitch them in Photoshop...

will give this a go while I wait for vray to implement a one click solution to 360 stereo renders.

edit: so that stitching program is actually quite useful! Photoshop would fail to find points in the slice to stitch. but what I do now is take two slices into Photoshop to figure out the amount of overlap that I have from my slice crop. You an dial this number into the StereoPhoto Maker software and get quite a nice image. I'll keep testing and share some images.

thanks!

this seems to work quite nicely!

my default settings:
64mm IPD distance between cameras
76200mm (250 ft) convergence point


current process (slice rendering):
- setup spherical 360 fov camera
- rotate camera with 360 frames (around IPD)
- crop render section, including overlap when moving between frames
- render sequence.
- take two slices, find out pixel overlap in Photoshop with difference layer type
- assemble in StereoPhoto Maker (http://stereo.jpn.org/eng/stphmkr/ (using mosaic & plugging in the pixel overlap)
- clip off & align any extra overlap from the edges (photoshop, offset)

now do it again for the other eye

I just need to finish building a test scene to share images (currently testing with scenes from work), I want to test how this works with objects close & far away - and eventually animation with a static viewpoint.

There is a free iPhone app called MS Photosynth that lets you take pictures of your full spherical surroundings, then stitches them together to create a panoramic skybox. The farther away your surroundings are, the better the stitching looks, and for the floor shots try to keep your legs out of frame, though some people use an all-directional mount. If you store your photosynth panorama online you can access it from photosynth.net. If your browser has Greasemonkey or Tampermonkey installed, you can install this userscript I wrote. It deploys a process to download the stitched image data, and also prepares VRML text that you can save to a file to map the panorama on a cube or a sphere. I hadn't considered depth maps for this though.

About nadir, zenith stereo distortions with 360 spherical stereo images ...

Here is a stereo equirectangular spherical image (360*180) in over/under format that has a very wide vertical coverage (shot with two Gopro Hero2s with 3d kit with the cameras in rolled orientation - about 30 degrees off vertical. This allows you to use the diagonal of the 1080p format on this camera to get the maximum vertical coverage out of the lenses -- about 160 degrees. You can view this in the Rift in VRPlayer (spherical, over/under).
http://www.stereopanoramas.com/rocksoverunder.jpg

If you look straight down 90 degrees to the nadir with this panorama in the Rift -- you see a black circle. If you are looking straight down though you will notice that the pavement details in the bottom half of the view appear to be behind the black circle in depth. If you now rotate your whole body 180 degrees, still looking down, you will see the bottom half of the view again appears to be behind the circle in depth. The bottom half of the view with this panorama always appears to be behind the circle no matter which direction you are standing. So it is not surprising that nadir and zenith views often look defective in stereo panoramas produced by stitching thin vertical slices. One way around it is to have a flat nadir area and have the stereo window depth set to the distance to the nadir -- or you can have a featureless sky for the zenith.

Btw -- if you keep turning your head so you are upside down you will see that the stereo is reversed (pseudoscopic) -- hold onto a chair for safety ;) . If this was a CG scene looking at it upside down would have a normal 3d impression

"nukemarine" wrote:

"cheerioboy" wrote:
Anyone know a good way to view these? I've been doing some tests with rendering 3d 360 still images. Since I believe it's not possible at the moment to create a pair of 360 renders, I'd be happy just being able to shift the views with the depth pass. Does the depth pass need to be a specific depth to generate the correct image?

I'm using vray with 3dsmax, in case someone want's to code me a special 3d camera that renders a pair of 360 degree images :)

Ok, now I understand sort of what a depthmap is. I looked at this image it first and wondered what the black and white image is doing. I did not realize VR Player had a depthmap option (or even what it does) under "Media - Effects". When I loaded your image, put in spherical projection, FOV 110, Horizontal offset 30, and depth map effect it looked like crap. Lots of distorted artifacts. Then someone posted that the depthmap had a slider adjustment. I found that under "Tools - Settings - Effects" and set the slider to 0.000 for a good alignment.

So if I understand correctly, a depth map can be created using two cameras which ends up looking like an odd black and white photo. At the same time, a 3rd camera in between them can be used to generate the panorama (cylinder or spherical). These two photos then can be used to create a fake stereographic 3D. The benefit of a depth map over a general two image S3D is that the Zenith and Nadir (top and bottom of spherical projections?) are not distorted.

Something I've also noticed. Changing the FOV and adjusting the Horizontal offset has an impact. For this image, it seemed to work best with a 90 FOV and about 60 horiz. offset.

I've attached my VR Player settings as a text file (save it as a json file) and the original render from OP. I'm pretty amazed at how well it works. Learning new stuff everyday it seems.

The depth map offset shouldn't be set to 0, that means there's no offset so you're basically looking a single 360 pano with no depth effect. The renderings of a pano & depth map are both from a single camera's view. The black & white depth map is then used in VR Player to distort the original pano (the white in the depth map gets shifted more than the black) - this helps to simulate the second view for the other eye. Shifting the max offset in a way is setting the distance between the eyes - so when you push it too far you start to get unwanted artifacts since you're really shifting the view with the depth map far more than is really possible. There just isn't enough info in that single pano to create a decent second view..

sorry, I hope some of that makes sense.

Yes, I figured that out a couple of days ago . I basically imagining depth just because the rendering had enough cues (very embarrassing mistake). When I did add offset, I began to see the issues that this method creates with small or thin objects. It's a cool idea, seems to work ok with up close large objects without a far off background, but is not something to strive to use. Your stereoscopic 3D renderings are still really good though.