Denoising Bottleneck with Mutual Information Maximization for Video Multimodal Fusion

38731

Knowledge Triples