This repository was archived by the owner on Sep 8, 2021. It is now read-only.
This repository was archived by the owner on Sep 8, 2021. It is now read-only.
Large values for saliency of uninitialized Inception_v3 #21
Hi,
There is something strange that happens when one used the torchvision inception_v3 model (pretrained=False), the resultant saliency map has very large values (order of 10^8).
Saliency Algorithm: Guided Back Propagation (from torchray.attribution.guided_backprop import guided_backprop)
Such randomization of weights helps to perform sanity checks.
Regards,
Asim