Code repository for the paper 'ViNNPruner: Visual Interactive Pruning for Deep Learning' (https://arxiv.org/abs/2205.15731).
An interactive demo can be found at https://interactive-pruning.dbvis.de/.
This is the home of ViNNPruner, an interactive pruning tool for deep neural networks.
In order to run the frontend, you need Node 12LTS or greater.
It may be possible that you might need to install vue-cli depending on your system configuration. If there are any errors during build, please also run:
npm install -g @vue/cli
Navigate to the frontend directory and run:
npm install
Afterwards, you need to build the project.
npn run build
After completion, run the frontend web server by entering the following command:
npm run serve
The interactive pruning backend uses PyTorch. It was originally written for PyTorch 1.8.1. It is highly recommended using an environment.
Navigate into the backend directory and run:
pip install -r requirements.txt
After all packages installed successfully, run:
flask run
The app will be available at http://localhost:8080
Results may take a while to load on first-run as datasets are downloaded on demand.
Used Third Party Libraries & Research
- Vue (MIT, https://github.com/vuejs/vue/blob/main/LICENSE)
- Vuetify (MIT, https://github.com/vuetifyjs/vuetify/blob/master/LICENSE.md)
- Flask (BSD Clause 3m, https://github.com/pallets/flask/blob/main/LICENSE.rst)
- Pillow (HPND License, https://github.com/python-pillow/Pillow/blob/main/LICENSE)
- Pytoch-Model-Summary (MIT, https://github.com/amarczew/pytorch_model_summary)
- Lookahead Pruning (https://github.com/alinlab/lookahead_pruning)
Released under MIT License. See the LICENSE file for details.
@conference{,
author = {Schlegel, Udo and Schiegg, Samuel and Keim, Daniel A.},
booktitle = {Machine Learning Methods in Visualisation for Big Data 2022 @ EuroVis 2022},
title = {ViNNPruner: Visual Interactive Pruning for Deep Learning},
year = {2022}
}