Hacker News Reader

Comment by EvanAnderson | Hacker News Reader

I've done this with my SLR. Moving the camera different amounts can give a more pronounced effect, however it can be more difficult to get the image to converge.

I had a lot of fun with doing this 20 years ago. Sadly, my visual acuity has become significantly different between my eyes (even w/ correction) and the enjoyment of 3D displays has really diminished as as result.

Just musing because I'm working and busy:

I wonder how difficult it would be to do video. (Obviously you'd have to shoot two videos in parallel versus just moving the camera and shooting again.)

Converting existing 3D videos to a cross-eyed viewing format would probably be the easiest way to experiment with it. I wonder if anybody has done that. I've never looked at 3D movie formats before. I always assumed it was two interleaved streams.

View on HN

I remember seeing video done this way on youtube during the 3D tv craze about 10 years ago (not the 3d stuff that youtube supported, this was just people messing about with 2 images side by side). It worked about as well as the examples here, but was not a particularly comfortable experience for anything longer than a few minutes.

For that matter, there are a couple of "Magic Eye" videos out there, which look like TV static until you cross your eyes enough.

> I wonder how difficult it would be to do video

One approach is time-shifted side-window video. Time-offset frame pairs from mono video that's pointed perpendicular to camera motion. Pretending that frames offset in space, time, and orientation, are offset only in horizontal-space-perpendicular-to-shot. Sufficiently perpendicular that difference in apparent size within pairs doesn't bite, and similarly for motion stability, orientation, distance from subjects, and non-horizontal motion. Phone video out the side window of a train/plane/bus/car with a view, or a drone footage segment that meets constraints. River seen from bridge-crossing train window; waiting rocket seen from circling drone. Upside is easy capture, and an inter-ocular distance that's potentially quite large (miles) and adjustable in post processing. Downsides include non-static things don't fuse (traffic by that river, fog off the rocket), and it can be a pain to find usable segments in existing video.

With AI generating depth maps from mono images, and filling in image gaps, it may someday be possible to generate stereo from some mono video. One challenge is visual artifacts involving depth can be very noticeable.

You should move it by the distance of your pupils for the best effect. Mine is bog standard 63mm so I'm lucky.

If you get it wrong things will look too big or too small and the 3D effect will be softer or too pronounced.

The point is that you can force the respective of a giant or an ant if you so choose.

Using two cameras I had a time lapse of a fog coming in where you could see the 3d structure becuase they were at two different streets looking out.

Like op this was done nearly 20 years ago. Now you don't even get fog any more.

I remember reading about a physical device someone constructed with mirrors (I think at an early burning man) that gave you the experience of a huge inter ocular distance to get a giant's eye view. I've always wanted to try one (or the opposite to experience a tiny inter ocular distance).

Found it http://eyestilts.com/intro.html

A pair of periscopes laid flat will do it, but it'll really confuse your eyes. I can feel it on these images: the eye muscles are trying to converge on a distance closer than the focusing muscles want to focus on, and I can tell that's a bit weird. That might be an age thing though.

You get the same thing with VR headsets as their physical focus is always around 1 metre away (3-4 feet). But you get used to it quickly.

The reason you need to change the inter-frame distance is because the amount of information carried by the parallax drops off quite quickly, to the extent that at the sort of distance in the tree photo, your brain is mostly using motion cues for 3d reconstruction, not stereo vision. Increasing the horizontal distance simulates bringing everything correspondingly closer.

Yes but that makes it look like a miniature world.

Reply on news.ycombinator.com