We assume batch_size = 1

:param x: [B x in_channels x T] :return: [B x out_dims x T]

:param x: 1 x D x T :return: 1 x D' x T'

Given a set of predicted actions probabilities A_{i}s, upsamples them w.r.t. the given projected L_{i}s. :param L: [K] The projected

:param Y: upsampled predicted probabilities [1 x C x T] :param A_hat: A^ set of actions in the video [M] each item between 0 and C-1 :re

:param A: :param L: :param V: :param A_hat: :param cfg: :param weight: :return:

Tries to use some heuristic to make the prediction the same size as the target. If the prediction is shorter, it will add background class at

converts predicted tensor of sets to python built-in set :param input_tensor: [C] :param thr: 0.0 <= thr <= 1.0 :return: \set\=M'

Does the reverse of summarize list. You give it a list of segment labels and their lengths and it returns the full labels for the full sequen