Dynmi/AlexNet

About loss

Closed this issue · 3 comments

Lxy85 commented

Hi,I used a face dataset with a total of 100 categories, each category is 5 pictures, but the loss has been hovering around 6 during the training process, and the batchsize and epoch are modified, but it still cannot be reduced.

Dynmi commented

Hi, Lyx85. Your dataset is so tiny that it couldn't make AlexNet work...

Dynmi commented

You can compare it with ImageNet...

Lxy85 commented

Thank you