DNNV

DNNV



Production


 The current production environment is hosted at the following url:

  https://dnnv-36899.firebaseapp.com/

Overview


  A web-based tool for visualizing neural network topologies(directed acyclic graph). It currently supports UC Berkeley's Caffe framework. DNNV is an extension of ethereon/netscope framework with added functionality. This application uses Firebase as database storage to store files and the users password and username.

Cloud Storage Functionality


  This application expands the Netscope project by allowing users to save their neural network topologies or just directed acyclic graphs on cloud storage via Firebase. Hence, a user of this application may create a user account to take advantage of such cloud storage capabilities. Acessing saved neural network topologies will proceed in a very natural manner via a login screen. One can also publicly share their respective neural network topologies with other users.

Editor View


  The Editor View is the main interface of the application. This is where the user can copy a prototxt file, encoding the information of the neural network as specified by the caffe format, into the text editor to view the directed acyclic graph. The text editor may also be collapsed and expanded at any time. An important feature of this application is that one can rearrange and drag the nodes/vertecies of the graph representation of the deep neural network, and these changes to the graph will be reflected in the prototext file.

  Hovering over a node will create a small window that shows a description of the node, and one can use the menu on the right to select multiple nodes and collapse them into a single group by hitting the group button. In an inverse manner, one can reverse the grouping of nodes by hitting the group button again. In order to change the name of a group, double click on the grouped nodes and change the name in the contextual pop up.

Why use DNNV?


  We have expanded upon the already existing framework Netscope, adding a multitude of functions that were not available before. Below are examples of features that we have added:


Example Features

Menu Navigation

There is now a menu located in the top right-hand corner of the Netscope window which allows the user navigate between all of the new features integrated into Netscope.

User Authentication

User Authentication using a username and password, Github or Google authentication. This allows the user to save and upload their own neural networks and keeps them secure, allowing only those who the user allows to view or edit the networks.

Saving to Firebase's Cloud

Secure upload to Google's Firebase cloud storage. This allows the user to save their neural network prototext file and access them later.

Sharing

Sharing is now available once the user logs in. The user can share their neural network with anyone else who has an account associated with this application.

File Browser

Now that there is integration with Firebase Storage, the user can access and modify previously saved files using an integrated file browser.

Grouping

Grouping nodes is now possible. Users can navigate to the menu an click grouping which allows them to select nodes that they wish to group together and collapse them into a single node.