Evaluating Capability of Deep Neural Networks for Image Classification via Information Plane

This code is a PyTorch implementation of our paper "Evaluating Capability of Deep Neural Networks for Image Classification via Information Plane" accepted by ECCV2018.

The code is run on the Tesla V-100.

Prerequisites

Python 3.6.9

PyTorch 1.2.0

Torchvision 0.5.0

Steps on Runing IB on CIFAR 10

python experiment/cnn_model_cifar10_train.py

Citation

If you find this code useful, please cite the following paper:

@inproceedings{cheng2018evaluating,
  title={Evaluating capability of deep neural networks for image classification via information plane},
  author={Cheng, Hao and Lian, Dongze and Gao, Shenghua and Geng, Yanlin},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={168--182},
  year={2018}
}