MInD

Code for MInD: Multimodal Information Disentanglement.

Experiments

While some methods may stand out when evaluated by specific metric on one dataset, only MInD exhibits consistently competitive performance across all datasets under all metrics.

MOSI & MOSEI

UR-FUNNY

Note that we use UR-FUNNY Version 1 here.

MIntRec

Citation

@article{dai2024mind,
  title={MInD: Improving Multimodal Sentiment Analysis via Multimodal Information Disentanglement},
  author={Dai, Weichen and Li, Xingyu and Hu, Pengbo and Wang, Zeyu and Qi, Ji and Peng, Jianlin and Zhou, Yi},
  journal={arXiv preprint arXiv:2401.11818},
  year={2024}
}