I am currently testing out models for the classifcation of images (in a multi label problem) and stumbled upon the implementation of a ML-GCN for that purpose in this paper: https://arxiv.org/pdf/1904.03582.pdf
Unfortunately, I can only find implementations in Pytorch and I was wondering how one would implement said idea with Tensorflow, especially regarding the GCN part. Also, in the paper, the model is used on a huge dataset with a least 100 different classes, some being highly co-dependent. For my purposes, I only have 20 classes in total that are generally not too dependent from each other. Would ML-GCN even a suitable idea or would a regular CNN likely perform better? Thanks in advance.