The Visual Revolution of the Vanishing of Ethan Carter

Photogrammetry is very common in film VFX, where I was first exposed to it nearly a decade ago. There is a very interesting technical merger taking place between photogrammetry, computer vision, and machine learning. My company, www.3D-Avatar-Store.com, is one such company where we've trained neural nets to perform the entire complicated photogrammetry process, and then relax it's technical constraints, allowing for less than ideal source imagery. This "technical merger" is creating neural nets that "know" a given class of object, and are able to reconstruct them given less than complete information. This is achieved by training the neural nets to already know about their class of object, enabling them to "recognize" when presented with one, and then fill in the missing information not present in the limited source photo(s).

At the 3D Avatar Store we've trained our neural nets to reconstruct human faces and heads. It is a photogrammetry process deep inside, but 95% of the process described by Andrzej Poznanski is automated. There is a manual step where one approves and/or corrects the identified facial feature outlines while generating an initial 3D reconstruction. Because we use a facial recognition neural net as a starting analysis, the jawline is hard to recover accurately (facial recognition ignores jawlines). So we have a "post reconstruction" series of deformation tools to allow the user to fine tune or exaggerate their reconstructed 3D form.

https://3d-avatar-store.com/A-peek-at-new-Fuse3DAvatars

And the best part of a process like this is all "3D Avatars" created by our process share the same topology: meaning they can be morphed between one another and controlled by the same animation rigs.

I love this "Visual Revolution of..." blog post. People need to know about photogrammetry, and the advances around it.