ABSTRACT
Note that if there is a single possible latent value for each example, then is linear in β, and we obtain linear SVMs as a special case of latent SVMs.
3 TRADITIONAL NON-MAXIMUM SUPPRESSION ALGORITHM
NMS is very important (Everingham M. et al., 2007) in the processing stage of the target detection algorithm. Reducing duplication detection can effectively improve detection precision. As the detector is used in all the scales and all positions, in the reserved detection window, there is inevitably a lot of duplicate detection, that is, a target for multiple initial detection window. Therefore, it is necessary to use NMS to further screen. Figure 2 gives the results of the suppression of duplicate detection.
