Make 3D models from your photographs with Autodesk 123D Catch

Autodesk’s 123D Catch software enables anyone to make a 3D model from a set of photographs. The model can then be manipulated or printed on a rapid prototyping machine (Image source: Autodesk)

A friend of mine recently told me about Autodesk 123D catch – new software with which one can create impressive 3D models of any object by just analyzing a set of photographs taken at different angles.The software is clever enough to figure out the camera positions so you don’t need any kind of special rig to make it work. However, for best results you should use consistent lighting and camera settings.

All you need to do is photograph the object from different angles – preferrably as many as possible. It is also a good idea to include some detail in the background so that the software can use this to determine the camera positions for the different photographs.

Eager to try it out, I put my Nikon D7000 with its huge 70-200mm F2.8 lens on a small wooden chair. I kept the tripod foot attached so the camera will  be somewhat raised from the platform and its underside visible. I then proceeded to take 50 photographs from different angles. Then I loaded them into 123D Catch and waited for the cloud to compute my model.

A collage of some of the images that I used to create my first 3D model

You can see my first result in this screen-grab video:

While not great looking, it worked well enough for a first try. Even parts of my living room (including a pair of slippers) were captured by the software! These background regions were not even in focus and partly occluded, yet ended up impressively identifyable.

Unfortunately the resulting model is very rough – especially near the deeply convex lens hood. Crisp detail (visible in the input images) are blurred away by non-overlapping regions and looks rather plasticky. It didn’t help that 19 of the 50 photographs were not identified properly and kept out of the final model, even though they were correctly focused and aimed. So there is a lot of room for improvement.

This software is very impressive, but making good models is still not a straightforward task. I suspect it will take a lot more practice and patience to get the hang of it.