this method takes the data input images and removes those that do not have a 3rd dimension to prevent issues with inception and returns the

Creates a graph from saved GraphDef file and returns a saver.

Returns the feature representation of some images by looking at the penultimate layer of inception-v3 Parameters ---------- X : ndarr

Returns an numpy array of the images in a folder directory Parameters ---------- Specie : string just the name of the class -

This function does training for the last layer of inception-v3. It either performs a cold start in which it starts from a pre-saved graph or

Inserts the operations we need to evaluate the accuracy of our results. Args: graph: Container for the existing model's Graph. Returns:

Adds a new softmax and fully-connected layer for training. We need to retrain the top layer to identify our new classes, so this function adds

Returns an index within the allTestFeatReps matrix for which is the closes to the target in feature spacee and belongs to the base class Par

Creates a graph from saved GraphDef file and returns a saver.

helper function doing the backward step in the FWD-BCKWD splitting algorithm

helper function doing the forward step in the FWD-BCKWD splitting algorithm

Returns the poison image and the difference between the poison and target in feature space. Parameters ---------- targetImg : ndarray

Returns the ids for the duplicates of training in test Parameters ---------- X_test : ndarray the feature represenatons of th

Returns the train-test splits of images and their feature representations. Parameters ---------- directory : string, optional