How to input a 3d model into ML algorithm?

I have a machine learning model that uses csv with measured data about buildings: width, length, height etc. I use it to predict some features and it works properly.

I would like to drop csv with length, height and width, and I would like to use some kind of algorithm to parse 3d model into the ML algorithm.

The second reason is to try this approach with nonrectangular buildings, which are hard to describe in simple csv generated by humans.

I do not necessarily need the algorithm to read the width, length etc. but to be able to predict some values based on training set of 3d models acompanied by csv with results.

What should I look into? Where can I find information about this approach?

Topic 3d-reconstruction data feature-selection machine-learning

Category Data Science

1
answered at

Convolutional Neural Networks can work with 3 dimensional data, and they don't require much feature engineering. Provided you have enough training data, that seems like a way to go. As long as your data is in the form of a grid it is actually easy-peasy. Just imagine an image-like pipeline, but of the form [x,y,z,n]. Where n is the number of examples. As opposed to [x,y,n] (black and white) or [x,y,c,n] (color-channels).

It might be the case however that your data is in a vector format (as tends to happen with building models). I do believe there are experiments with straight up graphs and deep learning, but my feeling is that you might be better rasterising your model in that case.

You will probably need to add some layers to compress the data concepts (conv and pools etc) and end with a fully connected layer of 1 node (since you are looking for a scalar).

About

Geeks Mental is a community that publishes articles and tutorials about Web, Android, Data Science, new techniques and Linux security.