Introduction - If you have any usage issues, please Google them yourself
VLAD can be understood as a compromise BOF and fisher vector is to do kmeans feature point cluster, and then a feature point nearest cluster center to replace the feature point, the loss of more information Fisher vector is the feature point modeling with GMM, GMM actually a cluster, but it is considered a feature point to the distance to each cluster center, which is a linear combination of all the cluster center to represent the feature point in the GMM process modeling is also a loss of information VLAD like BOF, as only consider feature points the nearest cluster center, VLAD stored for each feature point to the nearest cluster center distance the code is mainly used in video processing for use in extracting features VLAd encoding.