lallla1 opened this issue 5 years ago · 1 comments
Depending on how you would represent the image. You would potentially need to replace the graph convolutions with more traditional 2D euclidian convolutions, but can keep the siamese part.