search

yulunzhang / RCAN

PyTorch code for our ECCV 2018 paper "Image Super-Resolution Using Very Deep Residual Channel Attention Networks"
1.38k stars 314 forks source link

重头训练的过程中loss突然变大，这是什么原因呢？ #98

Open weishuoycbb opened 4 years ago

weishuoycbb commented 4 years ago

image

John-Yao commented 4 years ago

Hi, have you solved the problem? Thanks for you reply!

weishuoycbb commented 3 years ago

Hi, have you solved the problem? Thanks for you reply!

no

BarCodeReader commented 3 years ago

Hi, have you solved the problem? Thanks for you reply!

no

Hi,

do you mind to share your system configuration? I use 3090 and the latest pytorch & cuda to train RCAN, only 38.15 and 33.97 obtained on Set5 and Set14.

I also directly use set5 as test set during training(same as yours)...

Thus may I know your system configuration? since I feel the different version of libraries really hurts the performance.

Thanks a lot!

zejinwang commented 3 years ago

Hi, have you solved the problem? Thanks for you reply!

no

Hi,

do you mind to share your system configuration? I use 3090 and the latest pytorch & cuda to train RCAN, only 38.15 and 33.97 obtained on Set5 and Set14.

I also directly use set5 as test set during training(same as yours)...

Thus may I know your system configuration? since I feel the different version of libraries really hurts the performance.

Thanks a lot!

Hi, I encountered the problem of model non-convergence when training the RCAN model with an amplification factor of 2 from scratch, and the loss value has been oscillating around 1e29. Could you tell me how to deal with this problem?

raychl94 commented 1 year ago

你好，请问请问evaluateion 应该修改哪里的代码，把验证集改成 Set5 Set14 ......的数据集