I have been testing and comparing some interpretable techniques, such as Input times Gradient, Guided Backpropagation, LRP, and Integrated Gradients on breast cancer images and now I would like to compare numerically the explanations methods, which one produces quantitatively better heatmaps.
To compare the interpretable techniques I am using the innvestigate, in your opinion, It is possible to apply the pixel flipping algorithm with innvestigate? or Can you imagine how can I evaluate numerically each heatmaps using investigate?
Hello everyone,
I have been testing and comparing some interpretable techniques, such as Input times Gradient, Guided Backpropagation, LRP, and Integrated Gradients on breast cancer images and now I would like to compare numerically the explanations methods, which one produces quantitatively better heatmaps.
To compare the interpretable techniques I am using the innvestigate, in your opinion, It is possible to apply the pixel flipping algorithm with innvestigate? or Can you imagine how can I evaluate numerically each heatmaps using investigate?
Reference for pixel flipping algorithm: http://www.heatmapping.org/slides/2019_EMBC_3.pdf
Best Regards, Leonardo