source code and pre-trained model

[dwdai]

In 2021, i trained the basemodel and RL, and i got the similar result. But now, After the system of server is reinstalled，i retrain basemodel and RL , m@A and m@B decreased by four or five percentage points. if i used the based model that trained last year, i can still obtain the similar result, I really can't find the reason.

we guess it may be the reason for the pretrained Inception v3, which is updated by model library of pytorch.

if you have the inception v3 model that you used, please send me ([email protected]) , thank you very much