More details about individual files can be found in the readme file in the zip file.
Required: Gurobi and Yalmip optimizers

MAIN FILES THAT SHOULD BE RUN FOR REVERSE ENGINEERING RESULTS
revengmodel1.m, revengmodel2.m, revengmodel3.m
Things to change in code: iteration number (for iter 1:100) 
Update max number from 100 to whatever number
Change mat files to be loaded: load model2NN2.mat

NOT ALL THE MAT FILES ACTUALLY HAVE A NEURAL NETWORK TRAINED ON THE DATA. IN THESE CASES, INSTEAD OF QUERYING THE NETWORK, THE CODE DIRECTLY TAKES FROM THE OUTPUT
MATRIX CENTERMAT. COMMENT OUT THE CORRESPONDING CODE LINES (LINE 14 OR 15) TO RUN IT.
All mat files available in folder "mat files"
These mat files use vertex detection from computer generated images of prisms.
Upon running the code, four numbers will be printed out each loop iteration. These four numbers correspond to the parameter estimations with each loop iteration 
corresponding to one data point used to reverse engineer the parameters. 
All results are stored in the array total. Plot a histogram of the four columns of total to observe distribution of estimates.

There are also corresponding mat files for when only 6 of the 8 vertices of the prism available, corresponding to a more realistic setting when gathering actual 
data. This is in model1NN2.mat and model2NN3.mat. 
Currently some progress is being done on implementing the reverse engineering for real images. These images can be found in a folder called "real iamges." Note 
that some of the images are missing and moved into a folder called green_images in order to color the vertices green to experiment with automatic corner detection
using more vibrant colors. None of the code directly references either of these images. 
Mat file model2NNreal.mat consists of the same mat file format as before, but uses real data instead. The code runs with this mat file but does not produce very 
accurate or consistent results.

The structures of the neural networks used in the mat files are a combination of linear layers and leaky ReLU layers. See the readme file in the zip for more 
information on which .m files can be edited to change this.