Making 3D models from photos

[drkim]

There is some free software (now beta) that generates a 3D model from a series of 2D photos. I've tried it on a little plastic model, and the big Vasquez Rocks in California (you've seen them on Star Trek.)

It did a pretty good job. It didn't handle shiny surfaces very well, which is understandable because it sees the details in the reflections as 'behind' the object.

It also allows for manual adjustment of photos when it can't figure out how to correlate the features in them.

It's called 123D Catch:
http://www.123dapp.com/catch

[FingerFlinger]

This looks cool. Does it work sort of like a profiling 3D scanner? How many photos and angles does the software need?

[drkim]

No, a profiling 3D scanner is actually extracting 3D information, this works completely from a series of 2D photos. It uses the background as well as foreground image in the photo to segregate into 3D. (For example, you can't keep the camera on sticks and rotate the object, you must keep the object still and move the camera around it, and include the BG.)

They suggest at least 50 photos and no more than 70. The thing that hung me up was shiny stuff.

Hey, it's free! Try it out. If you are doing 3D gaming that lets you import your own models, it's pretty cool. And I don't know if/when they will start charging for it.

Here's their "how to" video:
http://www.youtube.com/watch?feature=pl ... TfXXJxDsXw

[cybereality]

Very cool. I will have to check this out.

I wonder if the models it produces would be suitable for a game engine, I imagine the meshes would be pretty dense.

[FingerFlinger]

Seems like it could be a good starting point for a modeler, or at least a relatively quick way to build up placeholder assets for prototyping.

[drkim]

@cybereality: It lets you select 1 of 3 different mesh densities:

• Mobile (low)
  Med
  High

@Finger: It's pretty high quality. This model was done without any interaction on my part, other than shooting the photos. The software auto texture-maps it as well.

Here is the low mesh:



This is the highest mesh:



...and here is the final object with texture mapped:

It's a little plastic figure, 2.5 inches high. I put the newspaper under it so the software would have a spatial reference. (It's easy to clip the 'floor' out of the mesh later.)

[FingerFlinger]

Wow, that's much better than I was anticipating! I've got a friend who is a sculptor; I'll have to show this to him.

[cybereality]

Yeah, that looks pretty good.

[drkim]

Here is a screenie from my Vasquez Rocks test. As you can see, it's not that great since I was unable to get above the rocks for the down angle, so a good deal of the upper textures are missing. Also, I wasn't being very organized about it, just wandering around shooting. If you keep the model in the same low perspective as I was walking around it, however, you don't see those missing tops.

This one also took some hand input to tell the software where some corresponding points were between photos.

However, it's not too bad considering how huge the thing is (those cars are in the FG) and it does wrap around the front 270 deg. (Which you can't see in this still - of course!)

At the bottom of the interface you can see a 'filmstrip' of some of my stills that went into the making of the model. Those little white cameras are generated automatically by the software, and actually pinpoint where you were when you took the picture.



It would be interesting to feed something like all the photographs of the Kennedy assassination into it. Not only would you get a 3D model of the plaza, but you could see where each camera was positioned at the time it took the picture.

[brantlew]

That little figurine looks great. What a cool little tool.