I just got some new prints from a few of my 3D photos taken on the Lume Pad. I have concerns about one of them.
The picture is of a mirror that has words written on it in a wine bar. Looking at the picture on the Lume Pad, it’s pretty easy to see that the words are written on the glass and that one can look deeper into the picture to see the counter and wine bottles in the reflection.
The print, however, has the words kind of floating behind the mirror, as if they were floating in air between the mirror and the counter. Additionally, they don’t appear to be at the same depth along the surface of the mirror. On one side, I even see artifacts where the text is stretching toward the front at the frame of the mirror.
It looks like instead of using my 3D photo, a 2D version was AI-converted to have 3D. I can’t imagine that would be the case, so I’m trying to figure out why it doesn’t look the same.
Photos weren’t good. Between the lenticular layer and not having enough separation between the two lenses, the photos are pretty unconvincing of anything.
LeiaPix Prints are very dependent on the photo and the quality of the depth map we can get.
They absolutely do not force a 2D to 3D conversion. They use stereo images when available.
With the right input content, LeiaPix Prints are the best photo-to-lenticular print I’ve ever seen. I spent over a year finding the right partner for this.
Unfortunately, not every image is going to work perfectly. If you’re unsatisfied with the quality of the image, please reach out to support.
In every LeiaPix Print I’ve done personally, if it looks good in 4V it looks great as a Print. That said, I’m exclusively using content shot with Leia devices. Our algorithm is tuned for that. Images from other cameras may have variable results.
Thanks for the info, @Nima! It may be somehow related to my contact lenses then. I get some sort of chromatic aberration with these multifocal lenses, and that can shift the perceived depth of something depending on its color. In flat, 2D art, orange and red items tend to appear floating on the page. It’s a neat effect, albeit annoying when one is trying to analyze 3D-ness.
I’ll take another look the next time I have glasses on instead.
I did some testing with my husband and he figured out the issue.
When looking at the picture on the Lume Pad, I get roughly 3 stereoscopic views across the horizontal plane. Given that it’s a 4V rendering, I’ll refer to them as 1-2-3-4 across the screen.
When looking on the Lume Pad at the center pair (2-3), the picture is correct. The words are shown flat across the glass. However, if I tilt (yaw, technically) the tablet so I’m seeing either 1-2 or 3-4, the words are behind the glass and at slightly random depths. (4-1, when you go too far past the main viewing cone, just looks wrong, so it’s easy to disregard that pair.)
If I were to guess at this point, I’d say that A) the AI generated views aren’t as good for handling words that seem to be hanging in the air (no surprise there), but perhaps more importantly B) the views used for the print were not the two views captured by the Lume Pad’s cameras. Again, still a guess, I’d venture to say that the print is using either one camera-taken view and one AI-generated view, and that’s giving rise to this anomaly.
The outcome of lenticular prints is pretty mouch all about the source material, in general You need a pretty wide stereo base or interocular distance to get a descent 3d effect in a lenticular Print.
In general You need a wider Stereo base If You are farther away to the object, there are disparity calculators out there for stereoscopic Photos If You want to dig deeper into the topic.
If You have shot the Pictures on your LumePad or the Hydrogen, chances are that because of the relatively small stereo base, they are taken for the inner Views and that the other views are generated with a depth map, as You already pointed out.
Unfortunately Mirrors and bright reflections are problematic If it comes to AI, Depth map and Viewport generation as some research papers in the last few Years pointed out.
Since AI in general hast made huge steps recently, this should only be a matter of time, until then I would try to avoid harsh reflections and mirror surfaces, or use a 3d Camera with a wider stereo base, a Camera slider or a multi camera Rig to shoot such Pictures.
If You need the Picture for Your Business You could also extend the Views with Your own handmade or edited Depth Maps, but that’s usually a lot of Work.
I do appreciate the complexities involved. Personally, for the lenticular print, I would rather avoid using an AI-generated view if it’s not necessary.
It’s not for business. It’s simply a photo I took with the only 3D camera I had while on vacation, the Lume Pad. It looks great on there (IMO, others may certainly judge for themselves ), and I was hoping the print would present an identical viewing experience and I could just frame it and hang it up.
Unfortunately, it is necessary. LeiaPix Prints uses dozens of views. Unless you want to shoot with a rig with that many cameras, the views have to come from somewhere Our AI does really, really good a lot of the time (especially compared to other places you can get automated 3D image prints from). Unfortunately, just like most machines, you can “trick” it with things that are very rare in the world, for example, floating text on a mirror. Images of food, pets, people, landscapes, etc. should work fantastically.
You could seek out a professional that does custom lenticular prints, and can take the original image you made and craft a more accurate lenticular print from it using PhotoShop and hand-alignment. But I assure you, that path will cost multiple times what LeiaPix Prints does! I know, because I’ve tried
), and I was hoping the print would present an identical viewing experience and I could just frame it and hang it up.
Our AI does really, really good a lot of the time (especially compared to other places you can get automated 3D image prints from). Unfortunately, just like most machines, you can “trick” it with things that are very rare in the world, for example, floating text on a mirror. Images of food, pets, people, landscapes, etc. should work fantastically.