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

关于训练时的图片数量问题 #12

Open EchoXu98 opened 1 year ago

EchoXu98 commented 1 year ago

你好,我想请问下,我们用了800张lr的DIV2K图像作为训练。在训练过程在,在终端时显示的是如下所示的(数量)/4000吗? [Epoch 371] Learning rate: 2.50e-5 [400/4000] [L1: 0.0139]com:0.042292844504117966 50.9+1.5s [800/4000] [L1: 0.0140]com:0.1750539392232895 50.7+1.6s [1200/4000] [L1: 0.0138]com:0.0321304053068161 50.9+1.6s [1600/4000] [L1: 0.0135]com:0.05393579229712486 51.0+1.6s [2000/4000] [L1: 0.0136]com:0.07788106054067612 50.9+1.6s [2400/4000] [L1: 0.0137]com:0.14176936447620392 50.9+1.5s [2800/4000] [L1: 0.0137]com:0.04981883987784386 50.9+1.4s [3200/4000] [L1: 0.0138]com:0.04940667375922203 50.9+1.5s [3600/4000] [L1: 0.0138]com:0.0642247423529625 51.0+1.4s [4000/4000] [L1: 0.0139]com:0.08730699121952057 51.0+1.4s

Zj-BinXia commented 1 year ago

程序是根据test的interval来确定1个epoch钟需要的图像数量的