search

WangYueFt / rfs

367 stars 66 forks source link

Doubt regarding training settings #5

Closed hgoyal5 closed 4 years ago

hgoyal5 commented 4 years ago

So according to the paper the weight for distillation and classification loss is 0.5 and 0.5 but in the script provided it is 1 and 0.5 . Which one is better?

For tiered imagenet, Cifar-fs and fc100 datasets are the batch size kept same as 64?

Could you provide the scripts for the datasets other than mini-imagenet?

HobbitLong commented 4 years ago

@hgoyal5 ,

Fixed this legacy issue. Both a=0.5 and a=1.0 should have similar performance.

Training recipe for others: (1) For CIFAR-FS and FC100: --epochs 90 --lr_decay_epochs 45,60,75 (2) For tieredImageNet: --epochs 60 --lr_decay_epochs 30,40,50

Other parameters are the same as miniImageNet.