XAI framework implementation with captum(https://captum.ai/): model interpretation with pytorch example codes with CIFAR-10 dataset, which can be replaced by other dataset
- captum :
conda install captum -c pytorch
- torch / torchvision
- numpy
- matplotlib
run main.py or in terminal,
python main.py --XAI_method 'IntegratedGradient'
XAI method can be replaced by other XAI methods listed as follows: GradientShap, DeepLift, DeepLiftShap, IntegratedGradients, LayerConductance, NeuronConductance, NoiseTunnel
which will produce explained heatmap and original image accordingly