Conviz is an example of visualizing weights and output of convolutional layers. In this project I show how to add convolution kernels and output of convolutional layers to a collection that fetched and visualized after training is completed. This approach can be easily generalized to arbitrary number of layers.

It is based on the following tutorials:

• 3_NeuralNetworks
• 02_Convolutional_Neural_Network

MNIST data set that is shiped with Tensorflow.

Plotted convolution kernels and output of convolutional layers are saved to disk.

• Numpy
• Matplotlib
• Tensorflow