I found a bug that causes a crash when the provided x array in the CarliniWagnerL2 attack is of dtype float64. The float64 dtype is propagated to y by the util function get_or_guess_labels. Since x is casted to float32afterwards, this causes a type mismatch during optimization.
I trivially solved this by moving the cast operation upwards which does not have any other effect but prevents the above bug.
I found a bug that causes a crash when the provided
x
array in the CarliniWagnerL2 attack is of dtype float64. The float64 dtype is propagated toy
by the util functionget_or_guess_labels
. Sincex
is casted tofloat32
afterwards, this causes a type mismatch during optimization.I trivially solved this by moving the cast operation upwards which does not have any other effect but prevents the above bug.