xialipku / rescan Goto Github PK
View Code? Open in Web Editor NEWRecurrent Squeeze-and-Excitation Context Aggregation Net for Single Image Deraining
License: MIT License
Recurrent Squeeze-and-Excitation Context Aggregation Net for Single Image Deraining
License: MIT License
I have traind your network and got a model, but when I run python eval.py
, I would get a error message:'cuda out of memory'
, My GPU is titan xp. What should I do if I want to do test?
BTW, can you explain what does show.py
do?
我按照之前被关闭的issue里讲的dataset设置方式,将Rain800设置为train集和val集,Rain100H设置为test集,运行train.py后,运行show.py,但是showdir里的图片显示几乎没有derain效果。
请问我该怎么做,才能复现论文里的结果呢?
Thank you for your code.i am very interested this .i have run this code successfully.but my model test is not well.and my gpu is 1070 .it is just 8g.so my model is not good enough.can you upload your trained model.i would be very grateful if you could upload your model.
Is Pytorch>=4.1.0 written incorrectly? What should be the correct version of Pytorch?
Hi,
I want to run the code for a project where I am evaluating different deraining models. But I am unable to access the Google Drive link. Is it not supposed to be be publicly available?
Thanks and Regards
thanks for your job
how to load your pretrained model our_net_G.t7
您好,我想问问您的data文件是如何放置的,我在运行代码的时候运行不通,还有就是想问问预训练模型在哪里可以下载呀
前辈,您好!近段在学习您上传的代码,并作了以下实验:
实验1.训练集用R800里的700张图像,验证集用R800里的100张图像,测试集用R100H。 结果为: ssim-0.75 psnr-24.01
实验2.训练集用rain_data_train_Heavy-R100H里的1800张图像,验证集和测试集都用rain_heavy_test-R100H里的200张图像。 结果为: ssim-0.42 psnr-14.19 目前尝试了多次改动,效果都不太理想。。。
所有的数据格式都是左图为标签图,右图为雨图的拼接图像。在训练和测试过程中没有对代码做其它改动(我将您代码里的参数设置与您的论文进行了对照,都是一致的)。 但是,所得到的实验结果却和您论文里相差很大。不知是哪些步骤出了问题?请指点,谢谢!
Hello
Can you please provide a better usage instruction for the algorithm? I did these steps:
1- I created three folders:
../logdir/
../models/
../showdir/
2- I downloaded the RAIN800 dataset which has three folders inside a folder with name rain
:
rain/test_nature: 92 real rainy photographs.
rain/test_syn: 100 images.
rain/training: 700 images.
3- I changed the name of the last folder.
training ---> train
Then, I set data_dir = 'path/to/dataset/folder'
in settings.py file to rain
. for example, the algorithm addresses the train folder as rain/train
.
4- I intentionally disabled all the codes which are used for validation to avoid extra errors. (please specify the validation dataset.)
5- I triggered the training command as python main.py -a train
But stilll I encountered the following error:
INFO - train--loss0:0.004158 loss1:0.00411 loss2:0.004155 loss3:0.004152 ssim0:0.7887 ssim1:0.8015 ssim2:0.7991 ssim3:0.7981 lr:0.005 step:1092
Traceback (most recent call last):
File "main.py", line 227, in <module>
run_train_val(args.model)
File "main.py", line 170, in run_train_val
batch_t = next(dt_train)
File "C:\Users\ashkan\Anaconda3\envs\pytorch\lib\site-packages\torch\utils\data\dataloader.py", line 276, in __next__
raise StopIteration
StopIteration
Regards,
Ashkan.
hello Train and Val with rain and no rain images how to place
Hi, @ XiaLiPKU. Is there any publicly pretrained model?
您好,请问第三章中提出的结合了大气散射模型的雨模型在下面的方法上发挥了什么作用呢?
楼主你说的:Please replace TrainValDataset with TestDataset in main.py when you run testing.
Sorry for not refering that in README.md,在你源代码里面test的时候就是用的TestDataset呀,哪里来的更换一说
Hi man, great job! I'm trying to understand your code by following the paper. However I didn't get what is the purpose of SSIM class at all. Please, could you gimme some tip about it? Maybe by point out which part of the paper is related to this module. Thanks!
Your dataset on the web is combined with the output,I need raw data sets that have not been processed,thank you very much!
I don't understand the meaning of some parameters in settings.py : stage_num, depth, use_se, frame. Could you briefly explain the meaning of these parameters ? thank you!
大概在142行
还有就是关于执行python main.py -a test 的时候报错
RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0. Got 481 and 321 in dimension 2
后来发现运行dataset.py文件显示的测试数据集大小为(3,64,64) 训练数据集大小为(3,481,321)(ps:测试图像是我自己选的)在这里给大家说一下。。。。
解决方法是在第99行sample{‘O’:O}前一行加上
O=cv2.resize(O,(patch_size,patch_size))
就可以了
RuntimeError: The size of tensor a (241) must match the size of tensor b (240) at non-singleton dimension 3
I get this error when I try to run eval.py.
Does anyone get this error as well and know how to fix it?
Thanks!
the line 194 in train.py, Maybe the sess.step += 1 should be corrected into sess.step += sess.batch_size
i found the setting.batch_size is 64 not 1
dataset.py中O = np.transpose(img_pair[:, w:], (2, 0, 1))一句取的是合成图中右边的没有雨的那张图,是不是应该写成O = np.transpose(img_pair[:, :w], (2, 0, 1))?
Hi. I read the codes in model.py. When it came to forward method in class ConvGRU, I noticed that the update of h you wrote, quoted
h = (1 - z) * h + z * n
Shouldn't it be like h = (1 - z) * n + z * h
? I've been focused on CNN for a while and just begin to learn RNN and GRU. I wonder if there is any difference between the update of h in GRU and ConvGRU. I really appreciate it if you or anyone can help me out on this line of code.
您好,我想问问100h数据集中原先的100张测试图片是200张中的哪一些?还有请问您有其他paper(jorder)在200张图的测试结果吗?谢谢
when the test datasets were solved at 80th batch, I encountered the problem: out of memory, and I found when the eval.py running , memory usage is getting bigger. I don't know why that happened.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.