This is TensorFlow code for experiments on MNIST and CIFAR-10 bipolar morphological ResNet-like neural network (https://arxiv.org/abs/2009.07190).
mnist_morph.py - the main file for MNIST experiment
cifar10_morph.py - the main file for CIFAR-10 experiment
BipolarMorphologicalConv2D.py - the file with bipolar morphological convolutional layer implementation
resnet.py - the file with BM-ResNet network implementation
utils.py - the file with helper functions
The implementation of ResNet network is based on the code from Keras https://keras.io/examples/cifar10_resnet/
In pretrained_models folder we share:
mnist_ResNet20v2_model.22bm.h5 - ResNet for MNIST with 22 BM convolutional layers
cifar10_ResNet20v2_model.22bm.h5 - ResNet for CIFAR10 with 22 BM convolutional layers
cifar10_ResNet20v2_model.16bm.h5 - ResNet for CIFAR10 with 16 BM convolutional layers