Photogrammetry and VR for Comparing 2D and Immersive Linguistic Data Collection (Student Abstract)

The overarching goal of this work is to enable the collection language describing a wide variety objects viewed in virtual reality. We aim create full 3D models from small number ‘keyframe’ images found publicly available Grounded Language Dataset (GoLD) using photogrammetry. will then collect linguistic descriptions by placing our reality and having volunteers describe them. To evaluate impact immersion on objects, we intend apply contrastive learning perform grounded learning, compare collected (in GoLD) versus models.