keplr-io/quiver

Name: quiver

Owner: Keplr

Description: Interactive convnet features visualization for Keras

Created: 2016-11-13 22:35:44.0

Updated: 2018-01-19 13:31:12.0

Pushed: 2017-09-22 13:26:02.0

Homepage: https://keplr-io.github.io/quiver/

Size: 517

Language: JavaScript

GitHub Committers

User	Most Recent Commit	# Commits

Other Committers

User	Email	Most Recent Commit	# Commits

README

Quiver

Interactive convnet features visualization for Keras

The quiver workflow

Video Demo

1. Build your model in keras

   l = Model(...)
   

2. Launch the visualization dashboard with 1 line of code

   er_engine.server.launch(model, classes=['cat','dog'], input_folder='./imgs')
   

3. Explore layer activations on all the different images in your input folder.

Quickstart

Installation

pip install quiver_engine

If you want the latest version from the repo

pip install git+git://github.com/keplr-io/quiver.git

Usage

Take your keras model, launching Quiver is a one-liner.

from quiver_engine import server
server.launch(model)

This will launch the visualization at localhost:5000

Options

server.launch(
    model, # a Keras Model

    classes, # list of output classes from the model to present (if not specified 1000 ImageNet classes will be used)

    top, # number of top predictions to show in the gui (default 5)

    # where to store temporary files generatedby quiver (e.g. image files of layers)
    temp_folder='./tmp',

    # a folder where input images are stored
    input_folder='./',

    # the localhost port the dashboard is to be served on
    port=5000,
    # custom data mean
    mean=[123.568, 124.89, 111.56],
    # custom data standard deviation
    std=[52.85, 48.65, 51.56]
)

Development

Building from master

Check out this repository and run

uiver_engine
on setup.py develop

Building the Client

cd quiverboard
npm install
export QUIVER_URL=localhost:5000 # or whatever you set your port to be
npm start

Note this will run your web application with webpack and hot reloading. If you don't care about that, or are only in this section because pip install somehow failed for you, you should tell it to simply build the javascript files instead

npm run deploy:prod

Credits

• This is essentially an implementation of some ideas of deepvis and related works.
• A lot of the pre/pos/de processing code was taken from here and other writings of fchollet.
• The dashboard makes use of react-redux-starter-kit

Citing Quiver

{bianquiver,
tle={Quiver},
thor={Bian, Jake},
ar={2016},
blisher={GitHub},
wpublished={\url{https://github.com/keplr-io/quiver}},