First-person perspective? about dagnet HOT 1 CLOSED

Hi!

All the architectures in this field work from a top-down view of the scene mainly because it allows using a global/absolute reference system for the entire scene, making it easier for the network to extract useful motion features and patterns.

I think working from a first-person perspective is possible, but I have some doubts about the precision of the final results since this choice would require some tricky pre and post-processing steps involving perspective and geometry (and, for DAG-Net you also need to think of a way of mapping our 2D goals into some type of 3D clues).

As it is, DAG-Net doesn't support the first-person perspective, but feel free to fork the repo and start from our code :)

Alex

from dagnet.