EPIC Fields Marrying 3D Geometry and Video Understanding Supplementary Material Ahmad Darkhalil David Fouhey

In this supplementary material, we first describe the companion video that provides an overview of our dataset (Section 1) and then detail how the data was released (Section 2) along with taking stock of additional information specifically promised in the checklist (Section 3). Next, we provide additional details on the dataset construction (Section 4) and on the benchmarks (Section 5). We devote a final section (Section 6) to showing that the EPIC Fields pipeline could be applied to reconstructing videos from the Ego4D dataset. We provide a short video in the form of a trailer at https://youtu.be/RcacE26eObE. It allows to visually assess how challenging the reconstruction problem is and hints at how frame filtering helps. The video also illustrates how the new camera poses complement the existing semantic annotations for this dataset (hands and active objects), showcasing the potential of marrying 3D geometry and video understanding.