csdwren / prenet Goto Github PK

I made a dataset of my own with only rain charts and no rain charts and no rain tracks like Rain100 can I train it?

hello! thanks a lot for your code!!
in the RainTrainL dataset, where do rainstreak-xxx.png and rainregion-xxx.png come from? how to get them?
if i create a dataset on my own, and there are only norain and rain image pairs(without rainstreak and rainnregion images) is it possilble to train on this dataset? if not, what should i do?
thanks a lot!

Hi I would like to ask you to train 100 epoch on each dataset?

Hello, author! How long does it take to train PReNet and PReNet_r on your device?

Hi, your work is great.
I have a question to ask you if I only want to train the LSTM model in stages 2, 3, 4, 5, 6, and 7 . Do I just need to change recurrent_iter=2,3,4,5,6,7 in networks.py and train_PReNet.py?
I don't quite understand, please tell me, thanks

您好，请问rain12 和rain1400 的训练集和测试集分别是什么啊？我在云盘里好像没找到，这让我很困惑。

测试数据集的rainy_image是在哪里？网盘里没有

You have 4 .py training files (train_PReNet, train_PRN, train_PReNet_r, train_PRN_r), but you have 3 training datasets (RainTrainH, RainTrainL, and Rain12600). When filling in the path, is there any specific .py corresponding to a specific training set path, or are these three optional?

Hello, thank you for your contribution!
I would like to ask how to put three models of 100H, 100L and 12600 together to train a model? If you train together, will you still train for 100 epochs? Will it be a waste of time?
Looking forward to your reply, thanks.

你好，我在使用训练代码时，加载数据出错，提示为
OSError: Unable to open file (unable to open file: name = 'datasets/train/RainTrainH/train_target.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)
训练集使用的是RainTrainH，该怎么解决呢？

作者你好 请问最后处理视频的那个视频在哪里下载呢 谢谢

你好，我在使用训练代码时，加载数据出错，提示为
OSError: Unable to open file (unable to open file: name = 'datasets/train/RainTrainH/train_target.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0)
训练集使用的是RainTrainH，
命令行为 !python train_PReNet.py --preprocess True --save_path logs/Rain100L/PReNet --data_path datasets/train/RainTrainL
请问该怎么解决呢？

你好，你们有没有试过每个阶段的卷积层（除了convlstm层）参数不共享的话，会降低多少？还有感觉计算y通道的psnr和rgb通道psnr的数值相差有点大，然后不同论文又都混着用，就比较麻烦

Hi. Does Rain100L training dataset only has 200 images?
Rain100L in other paper seems to have 1800 images for training.
Does this problem affect the experimental results?

你好，能分享下你的rain100H训练好的模型吗？prenet和prn任意一个都行，我发现matlab计算过后的psnr会比scikit-learn计算的高大概1左右，谢谢

hello! thanks a lot for your wonderful code!
I want to try ur code on my own dataset, so i create a folder named 'mydatasets' and put it in 'datasets/train/', and put rain and norain images in this folder.

i run 'python train_PReNet.py --preprocess True save_path xxxx -- data_path datasets/train/mydatasets'
and error occurs as below:
Unable to open file (unable to open file: name = ' datasets/train/mydatasets/train_target.h5') no such file or directory

so, how to get the file train_target.h5? looking forward to ur response!!!

您好我用rain100H训练100个epoch后并且测试出来的指标只有PSNR=28.0568，似乎并没有达到论文中29.46这么高

定义的PReNet 和 PReNet_LSTM 两个网络结构没有区别？

可以给下论文上的两张真实的数据集图片吗？

I entered a 2048 * 2048 picture for test results showing insufficient memory but input small resolution pictures can be, I can ask me to test a big resolution image?

How to obtain different stages of the derain image, like the Fig.1 t=1,2,4 in your paper?
Is it the intermediate output of a complete network or is it the result of constructing different networks to experiment with?

Hello, thank you for your contribution. I have read your thesis, can you answer some questions?

1. Does Loss only use SSIM and not MSE? What happens if these two Loss are used at the same time?
2. About four networks. During training, are these 4 written separately when typing commands?
   For example to train PReNet, I just need to type

bash train_PReNet.sh

and I don't need to type the remaining three

bash train_PRN.sh
bash train_PReNet_r.sh
bash train_PRN_r.sh

Looking forward to your answer, thank you

有以下几个问题想请教一下作者：
1.请问为什么这个数据集中除了由于无雨图外还有雨痕和雨图，这两个是怎么得到的呢（这个数据集是怎么制作的）？
2.四种图片类型放在同一个文件夹下又是怎么分别读取的呢？
3.在根据sh文件按要求放置数据集后为什么报错UnboundLocalError: local variable 'input_train' referenced before assignment，代码中的input_train和out_train是怎么设置或者读取（得到）的？
非常感谢！期待您的回复

数据集怎么转换成h5格式啊，博主教一下吧，求求了

I can not find a pre-trained weight of rescan, do you have it? can you share a copy? thanks.

请问，文章中提到的损失函数递归监督是怎么实现的呀？没有思路

您好作者，请问文件夹 ./PReNet/logs/real/中的两组参数PReNet1.pth和PReNet2.pth是在哪个数据集上训练出来的？

作者你好，在network.py中上述四个网络结构，为什么输出的时候倒数第三行的结果要加上input呢？x = x + input，这个应该是已经derain后的结果，在加上输入的图，是不是多余了？

您好，在我查看了Rain100H和Rain100L的数据集后，知道是密度大和密度小两类，但是密度的大小是以什么为标准评判的呢，难道是添加噪声的程度吗。以50%为分界线？

Hello, could you share your preprocessing datasets? Thanks!

hello, This result is re-trained? I do not find the result in original paper .thanks

Hi, your work has inspired me a lot！But also some questions that bother me.
Train Test
RainTrainH Rain100H
RainTrainL Rain100L，Rain12
Rain12600 Rain1400

Is this correspondence correct? I need to train the models separately on the TrainH，TrainL，Rain12600 datasets, right?

Hi, thank you for your contribution!
Will the training set of Rain12600 be trained 100 epochs. Then tested on data set 1400?
Looking forward to your reply, thanks.

I tried to read the datasets, but error shows

  File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
  File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
  File "h5py/h5f.pyx", line 96, in h5py.h5f.open
OSError: Unable to open file (bad object header version number)

any suggestion ?

Why inference time with RESCAN is less than PRENET, not as described in the paper. I use GPU.