【Machine Learning】【Python】Chapter 4: Hard Negative Mining for Optimizing Training of SVM Models ---- "SVM Object Classification and Localization Detection"

Jun 25, 2017#Machine Learning141

AI Translation

This post is translated from Chinese into English through AI.View Original

AI-generated summary

The article discusses the use of hard negative mining to optimize training SVM models in machine learning. The author explains that by using the original training set to train a model and then predicting negative samples using sliding windows, the accuracy of the model can be improved. The article also mentions that the number of negative samples should be much larger than positive samples in the original data.

Latest code GitHub link: https://github.com/HansRen1024/SVM-classification-localization

Introduction#

By optimizing parameters through PSO, the accuracy of the trained model has improved by about 4%. However, this is still not satisfactory. Therefore, we will try to optimize the training of the SVM model through hard negative mining. The basic principle is to first train a model using the original training set, and then predict all negative samples using sliding windows. If the prediction result for a window is positive, the window will be added to the training set with a specified label of negative. It is important to note that in the original data, the number of negative samples should be much larger than the number of positive samples.