PiPi
PiPi
> > It solved thanks, Yes, my synthText path was wrong! > > > This is the whole error for both trainic15 and trainSyndata: ==== syn_images, syn_gh_label, syn_gah_label, syn_mask,__ =...
而且，我发现，到达一定的训练阶段，loss值会上升。想请问一下，这是怎么回事呢？
谢谢你的回复，我想请教一下你，就是前面阶段的loss降的蛮快的，但是后面阶段loss突然暴涨，这是模型导致的吗？
@freesouls 你训练21个epoch之后的loss会暴涨吗？
@freesouls 中间过程的loss是会降低的，但是训练达到21轮epoch的时候，loss又会上升，我也觉得很奇怪
@chenjun2hao 我也想知道你当时跑的环境，我的是anaconda python3.6.8，cuda10.0，pytorch1.0.1
@yyfanxing 你好，换了优化器之后，这个问题解决了吗？
I may have solved this problem, but I don’t know why. The number of GPUs in the sample code is 4, but I have only 1. So I modify the...
> @TangDL What is the difference between your answer and duxiangcheng's? > I have tried it. It works! But another problem shows up. My environment is Google Colab. > >...
I have met the same problem. Have you solved it？ thanks！