I've always preferred stereoscopic views and wasn't really happy with the flat images of my Ricoh Theta camera. Therefore, I tried to establish a workflow which allows me to generate stereo panoramas using a single Theta on a rotating panohead. Results aren't perfect, but for me its a major improvement over these flat monoscopic images. I've included links to two samples which can be viewed on the Gear VR in the Oculus Photos app. The first one is derived from three shots in 21 sec using the old Theta m15 camera. For the second image I used the new Theta S camera taking four shots at 90 degree angles. For me it's really fun to revisit nice places I've seen on my Gear VR.
how do you generate stereoscopic from a theta cam ? just taking to pictures is not a valid option, since you got the parallax valid only for the perpendicular shot. (if your turn left or right 90deg the cameras are aligne just one behind the other, and if you turn 180, you even get left and right inverted.) I was thinking taking 360 stereoscopic by putting back to back 2 theta, but this problem heavily limit the reason to bother taking 360pictures versus simply shooting 150 with 2 gopro.
I'm rotating the camera clockwise in a circle and take 4 shots at 90 degree angles. Looking at the single lens output I get 4 left-right image pairs. In the example shown the left view is always derived from the front lens, and the right view from the backside lens. What I now have to do is separate (unstitch) the images. Then the 4 left view images can be stitched to a monoscopic left view panorama and the right view images yield the right side panorama. Joining these two images gives the stereoscopic panorama. Due to the wide FOV of the fisheye lenses it is not really necessary to take 4 shots. Taking three images at 120 degree angles yields enough information to generate a stereopanorama (this is shown in one of the examples I have made available for download).
thanks, nice trick . But for me stitching (and in that case also unstitching) is pretty annoying since the advantage of using the theta cam was we have no need to stitch pictures. I am currently thinking to a way to slices the shots to get the data directly from the pictures. you would have a center shot (left eye) and then 3 additional shots around the center one (right eye).
Nice! I recently started doing the same thing. In my latest experiments, I have been taking 12 shots around the circle, rather than 4, in an attempt to reduce the size of jumps in near subjects at stitching boundaries.
I'm also figuring out the math to perform viewpoint correction, so the zenith and nadir would line up better during stitching. Did you have any trouble with zenith/nadir? It's hard for me to tell whether your nadir shot has stitching artifacts, or if that's how your scene floor really looks.
Looking at trees and buildings I was surprised how good zenith workes. Nadir has much more problems because the floor is so much closer. The only thing I did here was to remove the remnants of the tripod by using the content aware fill function in Photoshop.
"Annogyl" wrote: Looking at trees and buildings I was surprised how good zenith workes. Nadir has much more problems because the floor is so much closer. The only thing I did here was to remove the remnants of the tripod by using the content aware fill function in Photoshop.
For outdoor shots, the true zenith is often distant enough to be approximately infinity, so no viewpoint correction is needed.
Nadir, on the other hand, is very often much closer, as you say, and can be subject to significant stitching artifacts. That's why I'm working on a theory for model-based viewpoint correction. For interior architecture shots, both nadir and zenith can be significantly close, plus possibly the walls too. I'm also thinking of making a viewer that knows how to re-un-viewpoint-correct the final stitched images. Such a viewer might be able to improve the apparent depth of single monoscopic panoramas too, by incorporating, at the least, a planar near-nadir model.
