• Minimum Knowledge of CV and ML
    There do exists the minimum knowledge of computer vision and machine learning in order for us to have a basic, fluent, non-embarassing conversation with professionals. There's a broad span of topics in these two fields and we need a bunch of books to cover them all. This blog just walks through the very basic pipeline of the most famous algorithms.
  • Scene Classification 101
    We implement a classification system to solve the classical problem in computer vision, a final project that is accompanied with the vision course. Old, antique, out of date features are used to salute to classics in old times before the renaissance of deep learning.
  • Pedestrian Detection 101 using HOG
    We implement a pedestrian detection system to solve the classical problem in computer vision. Out of date features (HOG) are used as the representation features and fed into the SVM training to obtain a detector.
  • Optical Flow Estimation
    We use the Horn-Schunck method to estimate optical flow in a coarse-to-fine way. Implemented in Python with subtle sparse matrix manipulations.
  • Panoramic Mosaic Stitching
    We implement a system to combine a series of photographs into a panorama. The software will automatically align the photographs based on their overlap and relative positions and blend the resultant photos into a single seamless panorama.