In this talk, I will discuss various approaches to visual scene understanding and future state forecasting. In particular, I will summarize the following projects:

- Object detection (CVPR'17) and human pose estimation in 2d (ECCV'18) and 3d (WIP).

- Modeling uncertainty associated with object instance embeddings (ICLR'19).

- Forecasting future trajectories of moving, interacting objects given past trajectories (CVPR'19 submission) and past images (ICLR'19).