This talk will reflect upon the advent of deep learning and its impact on computer vision, especially some of the most fundamental and practically valuable areas such as image classification, object detection and segmentation, holistic scene understanding, human-pose estimation, face recognition and, image creation and editing. I will provide the intuition behind the pre-deep-learning era approaches to solve the aforementioned problems and contrast them with the contemporary deep-learning approaches to bring forward the differences and common takeaways that can be applicable to other domains as well. Specifically, I will focus on the feature learning abilities, use of context, transfer learning, multi-task learning, common compute pipeline and assimilation of large-scale datasets in deep-learning systems.