Coder Social home page Coder Social logo

Comments (21)

infrontofme avatar infrontofme commented on July 18, 2024 1

您使用提供的数据集利用UWGAN合成fake water images了吗?再用您合成的fake water images和air images组成训练对来训练UNet

from uwgan_uie.

DataXujing avatar DataXujing commented on July 18, 2024

OK,非常感谢

from uwgan_uie.

DataXujing avatar DataXujing commented on July 18, 2024

大佬,在训练UWGAN时需要type1和type2分别训练模型,还是混在一起训练有一个模型就够了?

from uwgan_uie.

infrontofme avatar infrontofme commented on July 18, 2024

需要分不同的水质,分别训练模型

from uwgan_uie.

DataXujing avatar DataXujing commented on July 18, 2024

好的,明白了,非常感谢

from uwgan_uie.

DataXujing avatar DataXujing commented on July 18, 2024

后边的UNet也是分水质训练的对吗?

from uwgan_uie.

infrontofme avatar infrontofme commented on July 18, 2024

分水质训练,也可以尝试几种水质混合在一起训练

from uwgan_uie.

18804601171 avatar 18804601171 commented on July 18, 2024

@DataXujing 请问,训练需要做哪些修改嘛

from uwgan_uie.

DataXujing avatar DataXujing commented on July 18, 2024

自己调整一些参数比如,训练的epoch, lr, 优化器及训练数据的路径,就可以训练了

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

你好,请问我用uwgan生成了type1 类型的水下图像,也就是偏绿的图像,然后组成数据集训练u-net,参数设置和文中一样,损失函数用了l1_l2损失,学习率没有用余弦改变那个,就是用的1e-4,处理图像那部分is-aug我没用,只是shuffle了,是不是这部分是关键呀,测试集用的realc,test_one_image那个程序出来的图像感觉就像上了一层蓝,test出来的图像更差,根本就看不见,还有棋盘效应。这个问题出现在哪里呢

from uwgan_uie.

infrontofme avatar infrontofme commented on July 18, 2024

@shaxiaozi123456 和你说的因素关系应该不大,realc是公开数据集包含的场景复杂,各类水质都有,你是所有测试图像都这样的结果?还是有几张比较差?
我们跑的结果总体还可以,对于一些严重色偏或者雾状效果太重,表现也有待改进。

至于棋盘效应,有可能是损失函数引起的,你可以参考论文中关于loss function分析的部分。

近期,我们会把项目的预训练权重文件和合成的数据集整理后进行开源~

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

我这里所有的图像表现都很差,差到不如不增强,甚至我用训练集的图片用来测试,效果还是差。

from uwgan_uie.

DataXujing avatar DataXujing commented on July 18, 2024

我分水质训练,参数和paper中设置的完全一样,出来的结果也很差。不知道是不是哪里出了问题。

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

我分水质训练,参数和paper中设置的完全一样,出来的结果也很差。不知道是不是哪里出了问题。

我也是哎,出来的图片根本看不清

from uwgan_uie.

infrontofme avatar infrontofme commented on July 18, 2024

@shaxiaozi123456 如果用训练集的图片用来测试,效果还很差,那应该是没有训练好,你可以把训练的loss曲线发出来一下,把质量差的图片结果贴上来一下;或者你仔细检查哪个步骤出了问题。

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

img_ori_3
img_ori_18
img_real_18
res_gen100_18
res_gen200_3

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

在test_one_image中测试的realc的数据和训练集的数据,realc的数据都是这种偏蓝色,训练集数据恢复的也不太好

from uwgan_uie.

infrontofme avatar infrontofme commented on July 18, 2024

in-air images:
air_000000_000
syn-fake images:
3 types
fake_000000_000 fake_000000_000 fake_000000_000

restored in-air images: l1+msssim loss
fake_000000_000_gen

water images:
18_img_
73_img_
87_img_
380_img_
515_img_
635_img_

restored water images:
18_img__gen
73_img__gen
87_img__gen
380_img__gen
515_img__gen
635_img__gen

realc包含的水下场景很丰富,这种方法并不是在所有的场景下都能work,这也是水下图像增强算法的挑战。基于deep-learning的方法依赖训练集的质量,在模式比较接近的场景能够起到较好的效果。我们也正在做一些改进的工作,想要兼顾实时和较好的效果,拟融合一些先验知识到深度学习网络里,欢迎交流!

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

假如我全用绿的那种type,那么测试集偏绿的效果自然会好一些,如果混合三种type,是不是需要选择性的将某些空中图片对应水下图片,比如说空中图片相近的几张都类似,所以把这几张类似的图片分别对应不同type的水下图片,这样训练的模型泛华性能是不是应该好些,你们是怎么选择的,你们的实验的确做的很棒,就是我复现一个相对好的结果有点困难

from uwgan_uie.

infrontofme avatar infrontofme commented on July 18, 2024

使用UWGAN合成的水下数据集已上传到Google Drive,

fake water images generated from UWGAN: [Google Drive]

from uwgan_uie.

shaxiaozi123456 avatar shaxiaozi123456 commented on July 18, 2024

能上传到百度云吗

from uwgan_uie.

Related Issues (19)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.