Skip to content

eric8he/openmax-demo

Repository files navigation

openmax-demo

An end-to-end implementation of "Towards Open Set Deep Networks" (Bendale, Boult 2015). Existing implementations are not well-documented and are confusing to read, something that is fixed here.

Files included

generate_arrs.py - lays groundwork

openmax.py - the main event

templates/home.html - the frontend

static/styles/style.css - the frontend frosting

test-images/ - three images used to test in video/slideshow

Running the project

To start the Flask webserver, simply run

python openmax.py

Then the application will be available at http://localhost:8080/. If port 8080 is occupied, you can edit the SERV_PORT constant found at the top of the code.

This folder comes pre-loaded with a pre-calculated cifar10_dists.pickle file. If you wish to recalculate the MAVs and models for any reason, from the root directory, simply first run

python generate_arrs.py

to generate the needed pickle file. This will take upwards of 10 minutes, depending on your computer specifications. Then remove the old cifar10_dists.pickle and rename array.pickle to cifar10_dists.pickle.

About

Barebones end-to-end demonstration of the OpenMax open-set classification algorithm.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published