Code for reproducing experiments in "Understanding Impacts of High-Order Loss Approximations and Features in Deep Learning Interpretation".
- Python, NumPy, Pytorch, Argparse, Matplotlib
- A recent NVIDIA GPU
To evaluate the interpretation with default parameters on the given toy image, run python main.py. To access all the parameters use python main.py --help.
To generate the following examples use python main.py --lambda1 LAMBDA
Low values of λ1 lead to sparse saliency maps. Noise is removed and salient features are highlighted.To generate the CAFO example, use python .\main.py --image_path=IMAGE_NAME --lambda1=0 --magnitude
For the CASO example, use python .\main.py --image_path==IMAGE_NAME --lambda1=0 --magnitude --second-order