3D for the real world

When you scan a document, it produces a two-dimensional image. This is producing a traversable three dimensional space. It's a very different problem for so many reasons, and I don't think it's reasonable to expect the same fidelity in the result in the near future.

I too was confused but then I realised this isn't meant for fun hacky 3D-mapping. It's meant to sell stuff for the home and I think that's unfortunately going to limit the reach of this technology. The only third party type of business that might be able to make use of this is a real estate company for obvious reasons. I guess they have a public API that can pull the data but even then that requires somebody with the camera to upload your house.

I highly doubt IKEA or any other furniture store is going to bring their camera out to me just so I can then digitally place a few sofas and I'm not going to pay $500 just to... what... save me a trip to the returns department? If I bought it what happens when I'm done decorating? Does my camera become a useless $500 paperweight?

It's cool that they allow you to have an OBJ (UV maps and textures too?) - I assume that doesn't include whatever proprietary meta-data gets attached to denote walls, floors, et al. If I could get the mesh with just the camera I would buy one of these in a heartbeat but otherwise I don't see this being used by any consumer apart from the absurdly obsessive renovator.

Yes. The thing that sticks out the most to me is the poor treatment of edges. One possible reason the edges look bad is that meshing algorithms that take into account sharp edges can produce unrealistic mesh topologies that are difficult to work with. Bad mesh topologies can wreak havoc on some operations like texture mapping, which they obviously use quite heavily. There are algorithms that can clean this up, but it is a messy and computationally intensive business.

Or it could just be poor data from the scanner -- that would be unfortunate. Some of the artifacts are so bad that it would seem to suggest that this is the case (the corner of the countertop in the kitchen is quite bad, or the protrusion in the wall above the entry-way into the kitchen, for some examples). It's hard to tell from looking at the processed result. I wonder what the raw scans look like.

Very cool product though, even if it's rough right now. I can see this looking nice in the future.

Yes, and since everything is triangulated too early (it seems) there's no level-of-detail. The scan in the video looks like puke. The scan that plays in the background when you load the page first looks much better.

http://pointclouds.org/ has a subproject called (http://pointclouds.org/documentation/tutorials/using_kinfu_l...) that you can feed with a kinect or a similar device. It's quite fun!

If you have a kinect with a usb adapter you can try out microsoft's kinect sdk. There are a lot example apps including several 3d scanning ones (called kinectfusion I think?).