ranahanocka / MeshCNN

Convolutional Neural Network for 3D meshes in PyTorch
MIT License
1.56k stars 314 forks source link

How to use the classifaction model after training to classify a specific mesh #90

Open theveloped opened 3 years ago

theveloped commented 3 years ago

First off all thank you for the amazing work! I'm currently looking into classifying standard component (e.g. nuts and bolts) and this seems like a very nice approach for this. I was initially looking into using 2D views of the models with a standard CNN but this seems a lot more true to the real geometric information.

Now I have been doing some initial tests and it all seems to perform pretty well. I'm not exactly sure how to actually classify a mesh using a trained model. Do you have any example of doing this? Ideally I would like to have a list of the likelyhood per class.

ranahanocka commented 3 years ago

Hi @theveloped ,

This repo is doing supervised classification, so it requires a training set of your desired classes (e.g., nuts and bolts). If you don't have such a dataset, you might be able to use a pre-trained network, and extract the global feature vector for each shape in your dataset, and see if it can be e.g., clustered (or some other unsupervised technique), to see if it gives something meaningful.

-Rana