Abstract
Computational models are the means by which we test our ideas about how vision works. They have well-defined inputs, that bear some defined relation to the light impinging on the eyes. And they have well defined quantitative outputs, that relate to human judgements. Their success can be judged by how well they predict those human judgements, by how broad the range of conditions they encompass, and possibly by how well their internals match the neurophysiology. Tangentially, they may be judged by how useful they are in practical applications. In this talk I will share a few lessons that I have learned from my adventures in modeling of human vision.