Load all necessary model checkpoints from the checkpoint directory.

Assign seed to workers of dataloader. Make the results reproduceable.

clip delta, to make sure perturbation is bounded by [-eps, eps] and perturbed image is bounded by [0,1]

Return the function of clip image.

Starting from a given node, traverse downward through the condensed tree to find the first cluster where the

Given a list of vectors (vecs), this method finds the minimum norm element in the convex hull as min |u|_2 st. u = \sum c_i v

Compute loss for model. If both `labels` and `mask` are None, it degenerates to SimCLR unsupervised loss: https://arxiv.org/pdf/2002.0

r""" Overridden.

Add trigger to image. if batch==False, add trigger to single image of shape (C,H,W) else , add trigger to a batch of images of shape (

Setup padding so same spatial dimensions are returned All shapes (input/output) are ``(N, C, W, H)`` convention :param torch.Tensor

Returns a 2D Gaussian kernel array.

Return filtered poisoned dataset. Must call ABL.train() first to filter out poisoned samples.