Home > other >  Help: pytorch model of eval model far worse than the effect of "train" mode
Help: pytorch model of eval model far worse than the effect of "train" mode

Time:11-13

Have been validated not fitting, bosses, this problem has been the younger brother for a long time,

CodePudding user response:

I also encountered this problem, then solve the, because I didn't do the normalization of data, resulting in validation of bn layer and droupout the entire model, the influence of the data after normalization would not have this problem

CodePudding user response:

Data normalization, please refers to the training and test data sets using transforms. The Normalize (0.5, 0.5), I would do or test results the poor ah, cried
  • Related