Building Machines That See.docx

Building Machines That See Part I Summary Bottom-up processes can find structure in the visual scene including edges and depth cues • These processes can be modelled in computer vision by convolving a filter with the image and by matching distinctive features between frames Building Machines That Se...

Building Machines That See Part I Summary Bottom-up processes can find structure in the visual scene including edges and depth cues • These processes can be modelled in computer vision by convolving a filter with the image and by matching distinctive features between frames Building Machines That See Part II Summary Top-down processing uses knowledge about world and how things should appear to resolve perceptual ambiguity • Knowledge can be hard-wired but with deep learning technologies can now be acquired through learning • Bayes rule is an important tool for combining available evidence including prior expectations about the visual scene • Seeing the world as it is requires getting the right balance of bottomup and top-down constraints