My research focuses on machine vision, namely teaching machines to see. Our world is complex, three dimensional and dynamic. Computational models get to observe this world through images but only partially as visual data does not completely capture the depth and dynamism of the real world. The goal of my work is to design visual perception models that bridge the gap between 2D imagery and our 4D world. To this end, I build computational models capable of identifying objects and their positions from a single image, by estimating their 3D shape, pose, and 3D position. Additionally, I design models that can recognize humans, their actions, and their interactions with other objects. I am a proponent of open-source initiatives and have developed and led the design of renowned computer vision platforms, such as Detectron for object recognition and PyTorch3D for 3D deep learning.