search

Zj-BinXia / ENLCA

This project is official implementation of 'Efficient Non-Local Contrastive Attention for Image Super-Resolution', AAAI2022
86 stars 7 forks source link

为什么每次从断点处开始训练，学习率会在上一次的基础上减小一半？ #13

Open EchoXu98 opened 1 year ago

EchoXu98 commented 1 year ago

Continue from epoch 370... Making model... Total params: 41.22M Preparing loss function: 1.000 * L1 [Epoch 371] Learning rate: 2.50e-5

这是我训练时打印的部分信息，学习率应该在第400个epoch时减少至2.50e-5，可是这里继续epoch370后学习率直接成了2.50e-5

EchoXu98 commented 1 year ago

已经解决了