why the learning rate and batchsize in "train_imagenet.py" are different from what said in the paper? Which one can reproduce the result?And is the learning rate related to batchsize ?
AIshenfeng opened this issue · comments
Sorry for the late reply.
Yes, you should adjust the learning rate according to batch size. The learning rate in train_imagenet.py
performs similarly to the one in the paper according to our experiments.