Vis-Quant

Visualizing DNN Quantization effect on Network.

Vis-quant gives feature map visualization for quantized network with weight quantization.

① Functionality1: Bit-width selection of weight quantization.

② Functionality2: Batchnorm shift-term correction with double, triple, .. terms.

Running

Clone or download this repository:

git clone https://github.com/HanByulKim/vis-quant

Install the dependencies:

npm install

Run vis-quant:

npm run dev

Navigate to localhost:5000.

If port:5000 doesn't work, change port and run again:

PORT=8000 npm run dev

Navigate to localhost:8000.

Citation

Our project is based on the codes of CNN-Explainer.

@article{wangCNNExplainerLearning2020,
  title = {{{CNN Explainer}}: {{Learning Convolutional Neural Networks}} with {{Interactive Visualization}}},
  shorttitle = {{{CNN Explainer}}},
  author = {Wang, Zijie J. and Turko, Robert and Shaikh, Omar and Park, Haekyu and Das, Nilaksh and Hohman, Fred and Kahng, Minsuk and Chau, Duen Horng},
  journal={IEEE Transactions on Visualization and Computer Graphics (TVCG)},
  year={2020},
  publisher={IEEE}
}

If you find our project helpful, please consider to cite our project

@article{VisQuant2021,
  title = {{{Vis-Quant}}: {{Visualizing DNN Quantization effect on Network.}},
  shorttitle = {{{Vis-Quant}}},
  author = {Han-Byul Kim and Euntae Choi},
  project={2021SNU_infovis},
  year={2021}
}

License

The software is available under the MIT License.