For computer vision without context (2D images standing alone), we have some nice solutions already, but I think that as long as we keep using the same methods, it will be insufficient for many purposes. Because the truth is that projection of images to a square, 2D grid, and given the complexity of lighting, put us at a situation where we have insufficient information.

And we are already seeing this being heavily developed in autonomous driving systems and others, but I feel like the biggest computer vision applications will require much more information than a 2d image can offer. Instead, recognising objects when you have 3d information seems much more reasonable to me.