andwizard
andwizard
> 请问您有ACDCnight1006张那个数据集的链接吗？谢谢 请问lowlight的训练数据要怎么得到啊？是和fog_image类似用data_make生成吗？没找到相关py文件啊
recovery loss 都没有参与训练，作者应该只是单纯的算出来展示一下
While I can get some better detect result by the pretrain model 
> Hi, > > may I ask which dataset you used for training? 用的是作者给出的那些数据集，不过要经过data_make.py处理
> > > Hi, > > > may I ask which dataset you used for training? > > > > > > 用的是作者给出的那些数据集，不过要经过data_make.py处理 > > 你好,請問是使用作者的代碼進行training嗎? 是的
> @andwizard 你圖是如何弄出來的,我等下跟你比較一下 训练数据嘛？是这样的 
> > I can't training my model to lower test_loss, which fluctuating around 10, and the result of detect is bad, all the pictures seem to have been covered with...
怎么说兄弟，你弄出来了吗
> 你們都是用rtx30的顯卡嗎? 我GTX 1080兄弟
> hi, I also encountered the same problem, have you solved it? I switch to the torch-geometric==2.6.1 and it works