Video can be represented in many forms. One natural way that has been well explored is extracting application specific features, and then using them in the application. Another natural alternative is to extract common features that can be easily used by many retrieval applications.
In this work, we highlight the importance of extracting generic features, and present a generalized model for using them in retrieval applications. Applications developed from our model includes logical turning point detection, suggesting a video based on video similarity, and remix candidate search. Additionally, popular retrieval techniques developed using our model is much faster than the state-of-art retrieval techniques with comparable accuracy.
