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

grid_size: int of the grid height and width return: pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/

:param logdir: Path to logs: E.g '/work/sagar/open_set_recognition/methods/ARPL/log/(12.03.2021_|_32.570)/' :return:

:param pretrain_path: Path to full DINO pretrained checkpoint as in https://github.com/facebookresearch/dino 'full_ckpt' :return: weight

Make two datasets the same length

归一化 :param vec: :return:

Gathers the stats from self.trainer.stats and passes them into self.log, as a list

Evaluate clustering metrics on two subsets of data, as defined by the mask 'mask' (Mask usually corresponding to `Old' and `New' classes in G

idx2lb：每一个顶点对应一个类 lb2idxs：每个类对应一个id

Evaluate clustering metrics on two subsets of data, as defined by the mask 'mask' (Mask usually corresponding to `Old' and `New' classes in G

Calculate clustering accuracy. Require scikit-learn installed First compute linear assignment on all data, then look at how good the accuracy

In this case, the test loader needs to have the labelled and unlabelled subsets of the training data

:param obj: Moco State Dict :param args: argsparse object with training classes :return: State dict compatable with standard resnet archi