A quick way to make a 3D-model from two photographs

In 2010, shortly after its release, I bought Fujifilm’s first consumer 3D camera, the FinePix Real 3D W1. The camera is provided with two lenses and two image sensors, and therefore allows simultaneous capture of a left- and a right eye photo. It is fun to use, and the 3D effect can be very convincing. But unfortunately the display options for 3D images are quite limited, compared to normal 2D photographs.

Recently I realized that it is possible to convert left- and right eye photos to actual 3D computer models that can be rotated using a mouse. For instance, a company called AgiSoft is the producer of StereoScan, a piece of software (free download) that can transform mpo-files directly from the Fujifilm 3D camera to 3D-models which can be saved as pdf-files, that can be read by, and interacted with using the latest versions of Adobe’s free Reader software. With a model open in Adobe Reader you can then, to some extent ‘drag’ the image around, and thus visualize the scene from different angles.

The steps involved in making a 3D-model essentially are:

1) Record two photos, corresponding to a left eye, and a right eye view of your scene. A 3D camera, such as the Fujifilm Real 3D provides a convenient way to do this, but it can also be done using any normal digital camera, or camera phone and the simple ‘cha-cha’ technique.

2) Make sure that the two photos are properly aligned, to allow 3D viewing. If a 3D camera is used, the alignment is probably fine, but if the photos were recorded with a normal camera or phone they need to be aligned. The free program Stereophotomaker by Masuji Suto is excellent for this. Once you are set up to do that, it takes less than a minute.

3) Generate the 3D-model, using StereoScan, and save to Adobe pdf-format.

4) Open in Adobe Reader, and enjoy.

I did some experiments with StereoScan 1.0.1, using 3D photos recorded with the Fujifilm Real 3D camera. StereoScan is simple to use, you basically load a set of left/right photos, hit the create model button, and select a quality level for the model. With the default parameters and the high quality level, it generally took about 30 seconds to generate a model, using the mpo-format files (2 x 10 Mega pixels) from the Fujifilm 3D camera.

It was fun to experiment with, and when you understand what is possible, and what it not, reasonable results can be obtained. The biggest limitation is a fairly obvious one when you think about it: the software cannot make a model for the parts of the scene that were not ‘seen’ by the camera, and it can’t make a good model of something that wasn’t seen by both lenses. So clearly, there will be no model for the back side of objects.

Also, scenes where a foreground object obscures part of the background, which is very common in 3D photography, are problematic, because the missing information about the background will become evident as soon as one starts to rotate the model. What worked best were scenes where the foreground does not obscure any of the background. Even here it is rather limited how much the model can be rotated, without the view losing its integrity. How much of a limitation that is depends on the type of scene. For instance, if you shoot down a narrow alley, you intuitively will not expect to be able to rotate the view very much, because physically that would correspond to bumping your head, or the camera into the wall. On the other hand, if you record the portrait of a person facing the camera, you may naturally be curious about what the person looks like in profile, but here you will find that the view becomes corrupted much sooner than you would like when you start rotating.

Links to some of the best models I obtained are provided below in pdf-format. To check them out, make sure that you have a recent version of the free Adobe Reader. Depending on your computer configuration, the pdf-file may open directly in Adobe Reader, or in an emulator provided by your web-browser.

1) If the pdf opens within Adobe Reader, you are in luck! If you see a photo, use the mouse to click and drag it, to see the the magic happen. If you do not see a photo, click the ‘enable all functions’ button, after which the photo should appear. If it dosn’t, you may need a never version of Adobe Reader.

2) If the file opens within an emulator provided by the browser, you will probably not be able to see the model in this way. Instead, right-click the link to a model below, and save it to your disc. Then open the file from within Adobe Reader. As above, you may need to click the ‘enable all functions’ button if it appears.



Leave a Reply