Model visual explanation using Clara Train (eg grad-cam)

I noticed the infer_saliency.py script in the covid19 classifier repo. Is this a “visual explainer” on the classification inference? (such as a grad-cam?)
If not, is it possible to implement grad-cam using Clara Train?
Thank you!

Yes, that’s right. It’s implementing the Grad-CAM approach (Selvaraju RR, Cogswell M, Das A, Vedantam R, Parikh D, Batra D. Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization. arXiv e-prints [Internet]. 2016 October 01, 2016:[arXiv:1610.02391 p.]).

1 Like

The saliency maps are returned in nii.gz map and look kind of weird - is this because the model producing these saliency maps (covid19 class) is insufficiently trained on my dataset or am I missing something else? Both itk-snap windows pointing to the same location in the volume, highlighting a clear covid19-related lesion.

I would recommend loading the saliency map as an overlay in itk-snap and to chose a jet colormap for display. Some adjustment of the contrast might be necessary. You should see saliency maps like this: