PyTorch Implementation on Paper [CVPR2021]Distilling Audio-Visual Knowledge by Compositional Contrastive Learning