The way I don't understand is that model at the time of training, to belong to the category of my image, such as cats, it can extract features for judgment, for the cat image, the types and features might be a lot of very miscellaneous, may be dogs, cows, horses, etc., how to establish the characteristics of the model? I don't think loss may fall down,

CodePudding user response:

Actually this question is very simple, using softmax classification, the cat is close to 1, the final score of the cat's score is close to zero,