Spatially Transformed Adversarial Examples with TensorFlow
Add the possibility to provide grad_op to the optimization routine stadv.optimization.lbfgs, preventing growth of the graph. Not providing grad_op to this optimization routine will result in a memory leak in case of repeated calls.
First release.
Fix calculation of flow_loss to match Eq. (4) from arXiv:1801.02612.
Previously the summation over p (for looping over all pixels) was done inside of the square root, and not outside.