We now live in an era of big data especially of big visual data consisting of photos and videos. Although the quantity of visual data speaks itself by lending hands for novel approaches to long standing computer vision problems (e.g., enabling the success of deep learning for object recognition), we must not forget that we live in a well-structured physical world. Our world has structures in its geometry, appearance, motion, the way these vary and so on. We are merely capturing a slice of these innate structures as an image or as a video. My research goal is to fully exploit such inherent structures induced by the actual physical world on the images and videos at hand to make more effective interpretations of them. In other words, I want to unveil and tap into the tight, small world, in the otherwise big bloated, visual data. In this talk, I will draw examples from my past and current research, particularly on your eyes, crowds, and materials, to provide a better insight into what I strive for in research.