[basic knowledge of deep learning - 37] solve the imbalance between positive and negative samples

Imbalance between positive and negative samples is a common problem in detection tasks , In target detection , A large number of samples are negative samples without targets , Only a few are positive samples containing targets , and Focal Loss It is to solve this problem that .

Focal Loss Purpose

• focal loss The main purpose of is to reduce the weight of easily classified samples , Make the model pay more attention to difficult samples , Prevent a large number of simple samples from dominating the optimization direction of the model .

Focal Loss Calculation method of

• Solve the imbalance in the number of samples , Give cross entropy a weight in cross entropy , That is, the proportion of the opposite class . thus , For a small number of samples , The weight is relatively large , For a large number of samples , The weight is relatively small .
• Solve the imbalance of difficult and easy samples , Intuitive, , For the more difficult samples , Give a big weight , For easy samples , Add a small weight . The specific calculation formula is as follows ：
  

Bloggers will continue to update some basic knowledge related to in-depth learning, as well as problems and insights encountered in work , Please pay attention if you like 、 give the thumbs-up 、 Collection .