Excellent, thank you for sharing. - It's a shame that your naming wasn't as catchy as "Wave Function Collapse." I'm sure that's the only reason it's more popular.
Very interesting and well explained, thoroughly enjoyed this! I jokingly compared it to the candy crush algorithm :-)...well...except much more complicated!
Hello Paull Merrell, I really like your algorithm and its research paper. Do you think it would be possible to use a neural network to carry out the task of constructing the output model, an AI capable of determining the adjacency constraints and randomly drawing one of the possible labels. And also could this AI model be more efficient in terms of execution time and memory costs?
Yes, I think you could use a neural network. I'm not sure that would be more efficient in terms of execution time and memory cost. The algorithm is already pretty efficient. The current algorithm is focused on small local adjacency constraints. A neural network could be very helpful for capturing more large-scale constraints.
Another comment. An algorithm that would add items to the model that cause its completion to be impossible later, would be called a ‘greedy’ algorithm. It would be doing the best thing it could do if it only knew the local area and current state of the model, but it could do better if it looked at the whole model, and say predicted what might happen in the future.
Marching squares or cubes is a method for volumetric data using isosurfaces. I think you're using the term as a description of the tileset, but you don't get that for free you still need an artist to manually create the tiles as I do. Keep in mind, I'm presenting research I did in 2007. It's not as simple as you might think to create a non-greedy algorithm. The problem is NP-complete which are notoriously difficult to solve.
