search

pyg-team / pytorch_geometric

Graph Neural Network Library for PyTorch
https://pyg.org
MIT License
21.18k stars 3.64k forks source link

How to use Spline-CNN as Generator? #332

Closed Felijasha closed 1 year ago

Felijasha commented 5 years ago

❓ Questions & Help

Hi :) I'm trying the last to weeks to get a GAN work with Spline-CNNs. The Discriminator works pretty well, but I don't get a Generator work :/ Do you have any idea for the structure of the Generator and the forward function? I looked at several papers and Github-repos, but I still don't get it. My actual problem is that I can't imagine how the input noise should look like. I can't use just a noise vector like in other models. Maybe I just stand in the tub :D

Best wishes :)

rusty1s commented 5 years ago

This depends on whether you want to generate point clouds or meshes. For point clouds, graph convolutions like SplineCNN should be promising (see here). This approach can be easily implemented in PyG. For meshes, the task is much harder and is usually tackled by mesh deformation. Another interesting approach is the AtlasNet. However, given that the mesh structure is unknown during generation, this task can not easily be performed with something like SplineCNN.

Felijasha commented 5 years ago

I will have a look at it thank you for your very fast answer. Actually I try to implement a Generator and a Discriminator for MNISTSuperpixel.