BookerDeWitt / MSBDN-DFF

The source code of CVPR 2020 paper "Multi-Scale Boosted Dehazing Network with Dense Feature Fusion"
334 stars 69 forks source link

How do I need to modify the code to change the data set from.H5 to.Jpg format? #13

Closed sym1998 closed 3 years ago

sym1998 commented 3 years ago

Please!

BookerDeWitt commented 3 years ago

You can read the data from .h5 file with "h5py" package. Then you can switch the numpy format data to the image.

dingyan1478 commented 3 years ago

Hello! Do you know how to do it? I didn't use h5 file for training, just used. JPG file for training, but the training result is very bad, the output image has a mosaic effect, do you know why? Or does this network need to use H5 file for training?

f-w-m commented 3 years ago

Hello! Do you know how to do it? I didn't use h5 file for training, just used. JPG file for training, but the training result is very bad, the output image has a mosaic effect, do you know why? Or does this network need to use H5 file for training?

Hello, I am very excited to see your work. Could you tell me how to use JPG files for training? @dingyan1478

dingyan1478 commented 3 years ago

Hello! Do you know how to do it? I didn't use h5 file for training, just used. JPG file for training, but the training result is very bad, the output image has a mosaic effect, do you know why? Or does this network need to use H5 file for training?

Hello, I am very excited to see your work. Could you tell me how to use JPG files for training? @dingyan1478

I just changed the data preprocessing part and wrote the preprocessing code for JPG files. The other parts use the code provided by the author. However, in the process of training, the mosaic effect appeared for the first time, and then the super parameters in the code were modified according to the paper, and the effect was good. I hope it can help you.

Grass-clj commented 1 year ago

你好!你知道怎么做吗?我没有使用 h5 文件进行训练,只是使用了。JPG文件进行训练,但训练结果很差,输出图像有马赛克效果,你知道为什么吗?还是这个网络需要使用H5文件进行训练?

您好,我很高兴看到您的工作。你能告诉我如何使用JPG文件进行培训吗?@dingyan1478

我只是更改了数据预处理部分,并为JPG文件编写了预处理代码。其他部分使用作者提供的代码。但是在训练的过程中,第一次出现了马赛克效果,然后根据论文修改了代码中的超级参数,效果不错。我希望它能帮助你。

您好,您的修改的预处理代码可以提供一下吗?如果不行那就祝您生活愉快。