search

diegovalsesia / gcdn

Deep Graph-Convolutional Image Denoising
MIT License
94 stars 27 forks source link

can you release PyTorch edition? #2

Open xiebinghua opened 5 years ago

diegovalsesia commented 5 years ago

Hi, unfortunately we do not have a Pytorch implementation at the moment.

xiebinghua commented 5 years ago

Could you talk about the difference between your graph convolution and the counterpart in paper《Semi-supervised classification with graph convolutional networks》. Sorry, I'm not clear about GCN, but I am interested in this structure.

xiebinghua commented 5 years ago

I want to know that this code is for your ICIP paper or your journal paper ?

diegovalsesia commented 5 years ago

The code is from the journal version. You can find all the details on the graph convolution on https://arxiv.org/abs/1907.08448. The main difference of the graph convolution that we use with respect to the GCN by Kipf et al. is that it is computes the aggregation weights in an adaptive way, i.e. as a function of differences of feature vectors, and this function is implemented with a small neural network.

xiebinghua commented 5 years ago

thank you very much.