helloimlixin/AGFVisualization

Official code for "Attribution Guided Factorization for Neural Networks Visualization"

AGF-Visualization

Official code for "Attribution Guided Factorization for Neural Networks Visualization"

Baseline methods are taken from link, and link.

Prerequisites

• python 3.6
• pytorch 1.1
• torchvision
• numpy
• scipy
• scikit-learn
• scikit-image
• Pillow
• matplotlib
• OpenCV
• tqdm
• h5py
• imageio
• tensorboard

Usage

1. Create a new conda environment:

   conda create -n agf_env python=3.6

2. Activate the created environment:

   conda activate agf_env

3. Update conda to keep track of the environment:

   conda update --all

4. Install the PyTorch library inside the conda environment just created per the PyTorch official website, for instance, using conda with cuda toolkit version 10.2 on a linux machine, we have the command to install the current Long-Term-Support (LTS) Version,

   conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch-lts

5. Install other required libraries if absent:

   pip install matplotlib
   pip install scikit-image
   pip install opencv-python

6. Also the code should be able to run out of the box in Google's Colab: .

Good luck!