MEEV: Body Mesh Estimation On Egocentric Video

This technical report introduces our solution, MEEV, proposed to the EgoBodyChallenge at ECCV 2022. Captured from head-mounted devices, the datasetconsists of human body shape and motion of interacting people. The EgoBodydataset has challenges such as occluded body or blurry image. In order toovercome the challenges, MEEV is designed to exploit multiscale features forrich spatial information. Besides, to overcome the limited size of dataset, themodel is pre-trained with the dataset aggregated 2D and 3D pose estimationdatasets. Achieving 82.30 for MPJPE and 92.93 for MPVPE, MEEV has won theEgoBody Challenge at ECCV 2022, which shows the effectiveness of the proposedmethod. The code is available at https://github.com/clovaai/meev