Process Documentation: Photogrammetry to 3D Printable File

With notes on computational photography

tywen kelly
4 min readNov 3, 2020


The printed model (left) and a rendering of the 3D model (right).
The printed model (left) and a rendering of the 3D model (right).

Photogrammetry is a technique that takes images and generates an accurate 3D model of the objects depicted in the images.

A simple way to describe photogrammetry is that it take a series of images taken from many angles of one object. The object can be a tree, a rock, a person, or, if you can capture enough images, the object can be a forest, a building, or a mountain. Photogrammetry software takes these images and finds where they overlap and discovers where similarities in image patterns emerge.

Some areas of overlap will change more between photos, which means that they are close to the camera. Objects further in the distance will not change as much between images. Using these areas of overlap the software starts to calculate the depth of the object, and then uses that depth to generate a 3D model.

Photogrammetry can be highly accurate and is often used to calculate the specific volume of real world objects. It is also used in real estate and construction industries to calculate specific land mass down to centimeter accuracy.

More and more it is being used to “copy” real life objects and convert them to a digital file format to be later printed and replicated on a 3D printer. This is that flow I hope to show in this video today.

As you’ll see this process is still very manual. Thankfully it can be done today with almost all free open-source software, making it a relatively accessible medium, accounting for access to a 3D printer in the first place.

I created this video to document the process I went through to make my scan and prepare it for 3D printing. I made the video public in case it should help anyone hoping to do the same thing.

Notes on Computational Photography

Part of the reason I made this video so long was to show the extreme processing of images. I import and export the images and models between multiple specified programs to achieve a simple looking object. I wanted to show how computationally intensive this technique is.

Photogrammetry is a technique that disintegrates images into pixels and places those pixels in 3D points and reconstructs them into a new object. This reconstruction reveals something in the image that otherwise is not conveyed.

Reconstruction is a theme of photogrammetry, and it is also the meta-theme of the broader genre of computational photography. Computational photography inherently relies on the reconstruction of large dataset of images to do its work. Neural filters have been trained from large databases of images that have been curated as “correct” images. Face tracking, noise reduction, image compression, object detection, and rotoscoping are getting more realistic because they are trained on databases of realistic images. Whenever any of these techniques are used they are invoking the power of reconstruction. In a way they are reconstructing the “shadows” and “ghosts” of the data it was trained on and applying them to the new image.

Reconstructing the cliff face from dozens of drone images.

Reconstruction is interesting because it has biological ties. Our minds already do it. Take the way we remember things. Contrary to the way we tend to think of memory in our day-to-day lives, our memories are not just sitting around for us to retrieve it. Every time we remember an event, we reconstruct the event anew. When we are done remembering we re-encode our memories with our most recent way we remembered it. Memories are less retrieval than they are reconstruction. The debacle of eyewitness testimonies in the legal system confirms this.

Photogrammetry has the remarkable ability to reconstruct things at proper proportions. This can fool us into thinking that photogrammetry is retrieval, just like we fool ourselves into thinking our memories are accurate. Photogrammetry is indeed reconstruction. And, with the new ability to 3D print these objects, this reconstruction is re-enacted and materialize. 3D printing becomes a way to duplicate and reconstruct the object in a new space. 3D printing has a powerful effect to displace a photogrammetry-derived object from its original site. Any technique that can change where a place “is” will be used as a site of power struggle, and is worth paying attention to.

I am left with a few questions with what the meaning of photogrammetry: What will happen with this power to displace place? Who will have that power and what will they use it for? What does it mean that everything is reconstructed and not retrieved? What happens when databases and material objects collide?