Hi, I've checked the 'get_model_pool' function in 'model.py' and found no unpooling layer. Can we just use strided convolution layer here? As the authors stated in "Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering", pooling the rearranged graph signal is analog to pooling a regular 1D signal. Can we treat your 'conv2' feature layer as the so-called rearranged graph signal? Thank you for your attention.
lixiang-ucas
commented
5 years ago
Hi, I've checked the 'get_model_pool' function in 'model.py' and found no unpooling layer. Can we just use strided convolution layer here? As the authors stated in "Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering", pooling the rearranged graph signal is analog to pooling a regular 1D signal. Can we treat your 'conv2' feature layer as the so-called rearranged graph signal? Thank you for your attention.