search

HCIILAB / Scene-Text-Removal

EnsNet: Ensconce Text in the Wild
123 stars 27 forks source link

Training settings #18

Open janzd opened 4 years ago

janzd commented 4 years ago

Are the default settings of parameters such as learning rate the parameters that you used for training? I cannot train the network for more than about 10 epochs as the loss becomes NaN by then.

zhangshuaitao commented 4 years ago

You can try increasing the batchsize.

GuangtaoLyu commented 2 years ago

You can try increasing the batchsize.

I increase the batchsize from 4 to 10. the loss still becomes NaN.

UsamaRehman001 commented 1 year ago

Are the default settings of parameters such as learning rate the parameters that you used for training? I cannot train the network for more than about 10 epochs as the loss becomes NaN by then.

Hey you sorted out your issue?