One can use this code to visualize DNNs (activations, filters, network structure).
Main goal of the toolbox is to visualize networks that solve image classifications tasks. Current version assumes to be given a model trained and saved using Keras: The Python Deep Learning library Development of this toolbox was motivated by the paper:
- Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson. Understanding neural networks through deep visualization Presented at the Deep Learning Workshop, International Conference on Machine Learning (ICML), 2015.
Example of a network visualization that classifies different shapes
The main window consists of activation maps for selected layer, the input and the structure of a network where the layers of interest can be chosen
Installation process is just installing packages listed in the requirements.txt
-
python virtualenv:
-
Install your virtualenv following The Hitchhiker’s Guide to Python, check that python version >=3.6 is set as primary interpreter
-
After activating the enviroment, run the command to install necessary libraries:
$pip install PyQt5 numpy scipy matplotlib h5py keras tensorflow
-
Install PyTorch.
$ conda install pytorch torchvision cuda80 -c soumith
Then run the PyTorch MNIST example, tpye
$ python main.py --framework=torch