Towards a Holistic Scene Understanding

Data Skeptic

Machine learning has been an essential tool for solving computer vision tasks such as image classification, object detection, instance recognition, and semantic segmentation, among others. The crux of machine learning approaches involves data. Training a machine requires enormous amounts of usable data. Why? Suppose you want to learn about monkeys and apes. Let's also assume you've never seen any monkeys or apes in your lifetime, until one day, someone shows you a picture of a monkey and an ape. It might be difficult to generalize from one picture and discern the differences between a monkey and an ape. If you saw perhaps 50 pictures of each species, you would have a greater chance of noticing that monkeys tend to be smaller than apes and that monkeys tend to have tails, whereas apes do not. Now if you saw thousands of pictures of both monkeys and apes, it might become very clear to you that the two are in fact, very different. For example, you might discover monkeys and apes have different nose structures, upper bodies, feet and so

Next Episodes