Multimodal Weekly 78: Long-Take Video Dataset and Flexible Mixture-of-Experts

​​​​​​In the 78th session of Multimodal Weekly, we have two exciting presentations on video dataset and multi-modality combination.

​​​​​✅ Tianwei Xiong will present the first long-take video dataset LVD-2M, which comprises 2 million long-take videos - each covering more than 10 seconds and annotated with temporally dense captions.

​​​​✅ Sukwon Yun will present Flex-MoE (Flexible Mixture-of-Experts), a new framework designed to flexibly incorporate arbitrary modality combinations while maintaining robustness to missing data.

Join the Multimodal Minds community to connect with the speakers!

Multimodal Weekly is organized by Twelve Labs, a startup building multimodal foundation models for video understanding. Learn more about Twelve Labs here: https://twelvelabs.io/