moco v2 (momentum contrast for unsupervised visual representation learning)

This model uses a dynamic dictionary and a momentum encoder to learn robust representations from unlabeled data.