(2024CVPR) MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding - View it on GitHub