search

IGITUGraz / WeatherDiffusion

Code for "Restoring Vision in Adverse Weather Conditions with Patch-Based Denoising Diffusion Models" [TPAMI 2023]
MIT License
310 stars 32 forks source link

About training #25

Open chenzean opened 11 months ago

chenzean commented 11 months ago

I am in the process of training my own dataset and found that after 200000 iterations, the loss does not show a decreasing trend. May I ask if this is normal?

jiangyuhangcn commented 9 months ago

I am in the process of training my own dataset and found that after 200000 iterations, the loss does not show a decreasing trend. May I ask if this is normal?

Hi！I have the same problem as you. Have you solved it? How long have you been training? Has loss converged？ thanks！

Rakesh-123-cryp commented 7 months ago

Change the lr scheduler if you're using, to a less aggressive one. and increase dataset size

qianduodu commented 5 months ago

我正在训练自己的数据集，发现经过 200000 次迭代后，损失并没有呈现下降趋势。请问这是否正常？

为什么我的损失是从十几万开始下降，下降的很慢，而且下降到十几就开始跳动这个原因是什么呢