EgoDTM: Towards 3D-Aware Egocentric Video-Language Pretraining