D2NN相关问题

[babypure]

有些部分表示看不懂，方便留下联系方式咨询吗，请问code里面的u0(input data?)和theta表示什么

[ghost]

你好，代码你现在看懂了吗

[babypure]

看懂了 但是网络不收敛

…

________________________________ From: wqh-cyber <[email protected]> Sent: 23 September 2019 04:05 To: Lyn-Wu/Lyn <[email protected]> Cc: Xuechun Wang <[email protected]>; Author <[email protected]> Subject: Re: [Lyn-Wu/Lyn] D2NN相关问题 (#2) 你好，代码你现在看懂了吗 ― You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%3Femail_source%3Dnotifications%26email_token%3DAMEGEBC55L5K6723HWJSJ23QLAW7LA5CNFSM4H6ZU2EKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7JWHTY%23issuecomment-533947343&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7Cc95b919ee014453891da08d73fd2e2c8%7C569df091b01340e386eebd9cb9e25814%7C0%7C1%7C637048047297000386&sdata=xB8%2BaF61lta8HikeSFh%2FcVfk%2FahdVG1bawYfb8puyXY%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBB2T32W7DWG2LJM64TQLAW7LANCNFSM4H6ZU2EA&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7Cc95b919ee014453891da08d73fd2e2c8%7C569df091b01340e386eebd9cb9e25814%7C0%7C1%7C637048047297000386&sdata=iRoYa3YQ4pLHzxoHpvMsLl2R53FYpVMoGewfK72Zm6M%3D&reserved=0>.

[ghost]

我很好奇这个代码中的衍射方程，有明显的问题 而且孔径的透过率函数H明显与衍射方程不一致

[babypure]

嗯 主要是我改了很多次参数都不收敛 认为就是网络架构的问题了

…

Sent from my iPhone On 24 Sep 2019, at 01:31, wqh-cyber <[email protected]<mailto:[email protected]>> wrote: 我很好奇这个代码中的衍射方程，有明显的问题 而且孔径的透过率函数H明显与衍射方程不一致 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%3Femail_source%3Dnotifications%26email_token%3DAMEGEBHH52RSZ6KCVOKQ3ODQLFNUXA5CNFSM4H6ZU2EKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7MVMXQ%23issuecomment-534337118&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C97ef64dd73ad44b7551d08d740867f54%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637048818727557460&sdata=zmx6pyUn4%2B4SzvG0JNPJT89Dfrqu0pbxaVjdl2fFiEM%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBDOBBXI7GRUD3WPHBLQLFNUXANCNFSM4H6ZU2EA&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C97ef64dd73ad44b7551d08d740867f54%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637048818727567446&sdata=3RfLev3NIG4Uspt7j5OcKFw6UMKof8vAT4yhXP%2Bl0ok%3D&reserved=0>.

[ghost]

好吧 我昨天花了5小时跑了1个epoch 训练的结果是损失0.0001 精度为1 然后我就不想再继续运行了

[ghost]

我现在在思考就算我得到了每一层的幅度图，我该如何3d打印

[babypure]

那应该还要学一些3d建模的东西 他这个方法是改振幅 原paper是改相位

…

Sent from my iPhone On 24 Sep 2019, at 02:00, wqh-cyber <[email protected]<mailto:[email protected]>> wrote: 我现在在思考就算我得到了每一层的幅度图，我该如何3d打印 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%3Femail_source%3Dnotifications%26email_token%3DAMEGEBGYR6J3ER2KGQXL2WDQLFRAZA5CNFSM4H6ZU2EKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7MWV6A%23issuecomment-534342392&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C213eff749c5f416c4d1c08d7408a8536%7C569df091b01340e386eebd9cb9e25814%7C0%7C1%7C637048836030090881&sdata=agkM8H8Gy3L%2BCq%2B%2BGNlKw6%2BgWeo1R%2FfhnmuHK70iei4%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBGL3CYDVE3GRYZMSATQLFRAZANCNFSM4H6ZU2EA&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C213eff749c5f416c4d1c08d7408a8536%7C569df091b01340e386eebd9cb9e25814%7C0%7C1%7C637048836030100873&sdata=0JQDWQIErVBApEbcfJ0YfqV3TonA91h11qZ32HEo5Wo%3D&reserved=0>.

[babypure]

我没研究3d打印的东西 你可以看看原paper的supplement以及3d建模方面饮用的paper 里面是3d打印的每个点的高度和相位有对应关系 也取决于材料

…

Sent from my iPhone On 24 Sep 2019, at 02:07, wqh-cyber <[email protected]<mailto:[email protected]>> wrote: 对的，目前我知道的是visualsfm+meshlab可以进行多视角图片的三维建模，但是计算机训练出来的幅度图不是多视角的，就很好奇原papper是如何打印的了 这篇文章你复现过吗 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%3Femail_source%3Dnotifications%26email_token%3DAMEGEBH2HYLHHZQTGXWFHV3QLFR37A5CNFSM4H6ZU2EKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7MW55I%23issuecomment-534343413&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C99add4dc8c6440e6115b08d7408b89bb%7C569df091b01340e386eebd9cb9e25814%7C0%7C1%7C637048840360572007&sdata=PNtSdcfmU0zUasvDtZSWDLfCQ7rAnPQmuA9%2B8ka9Sxg%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBAUTKXPOUXKNFGB6P3QLFR37ANCNFSM4H6ZU2EA&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C99add4dc8c6440e6115b08d7408b89bb%7C569df091b01340e386eebd9cb9e25814%7C0%7C1%7C637048840360572007&sdata=AgxcpQojMwi2%2Ffr0PtQkHS%2BqFlF%2BVvYe8SpbAfJmUrQ%3D&reserved=0>.

[ghost]

好的，谢谢

[babypure]

如果你复现了如何相位调制D2NN 麻烦分享下 谢谢～

…

Sent from my iPhone On 24 Sep 2019, at 02:15, wqh-cyber <[email protected]<mailto:[email protected]>> wrote: 好的，谢谢 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%3Femail_source%3Dnotifications%26email_token%3DAMEGEBESNSSJOJTBBHE7M73QLFSZNA5CNFSM4H6ZU2EKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7MXIXA%23issuecomment-534344796&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C0016433e21194cd2ac8b08d7408ca172%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637048845052961776&sdata=OtfXCOwJM%2FaYSpmNZi9cFw382h1YMPlggGhHq5C%2F1sc%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBAGKU3RTKB5PD5GY6LQLFSZNANCNFSM4H6ZU2EA&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C0016433e21194cd2ac8b08d7408ca172%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637048845052971772&sdata=Tve%2BrGMT97DjP4rEbqtffvxodXykN6QktHrQhofCSKg%3D&reserved=0>.

[ghost]

经过我的仔细思考，我现在对这个代码完全理解了 ，老哥你的电脑配置如何？我把batchsize改成了10的话1个epoch需要5个小时，如果batchsize是64，计算机直接死机。 另外 代码中有一行D=dLdL/(Nlmb) 我个人觉得这行多余了，无论是远场计算还是近场计算 都不要这一行

[babypure]

我什么参数都试过，就是不收敛。。你有微信吗 要不微信说？我的：xcw1123

…

________________________________ From: wqh-cyber <[email protected]> Sent: 24 September 2019 04:04 To: Lyn-Wu/Lyn <[email protected]> Cc: Xuechun Wang <[email protected]>; Author <[email protected]> Subject: Re: [Lyn-Wu/Lyn] D2NN相关问题 (#2) 经过我的仔细思考，我现在对这个代码完全理解了 ，老哥你的电脑配置如何？我把batchsize改成了10的话1个epoch需要5个小时，如果batchsize是64，计算机直接死机。 另外 代码中有一行D=dLdL/(Nlmb) 我个人觉得这行多余了，无论是远场计算还是近场计算 都不要这一行 ― You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%3Femail_source%3Dnotifications%26email_token%3DAMEGEBEHWDFSO5ZVXPN7JV3QLF7UFA5CNFSM4H6ZU2EKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7M46PI%23issuecomment-534368061&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C1178e2aa3f1e4204c02208d7409bee92%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637048910809746334&sdata=jPQgc6fwhqgZUJPYuAQgoAxbRBLJUsEjZJKy12HbxiY%3D&reserved=0>, or mute the thread<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBGZKIBPWSNREHTGPJ3QLF7UFANCNFSM4H6ZU2EA&data=02%7C01%7Cxuechun.wang%40qmul.ac.uk%7C1178e2aa3f1e4204c02208d7409bee92%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637048910809756328&sdata=4eBZdj47EeLGnzmOpflRG168OGVmEduNasccs7yXdAo%3D&reserved=0>.

[closest-git]

D2NN的pytorch实现参见https://github.com/closest-git/ONNet， ONNet实现了一些新的模型，并在迅速改进中，欢迎交流，讨论

[ghost]

衍射神经网络的远场计算公式和近场计算我印象中有一篇论文中提到过，你那个代码是纯振幅调制而且训练过程不收敛，最好改成纯相位调制。 还有就是可见光的衍射神经网络你有研究么？ 可以加我 QQ: 1205134271&nbsp; &nbsp;

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "Lyn-Wu/Lyn" ***@***.***&gt;; 发送时间:&nbsp;2021年5月1日(星期六) 中午11:14 ***@***.***&gt;; 抄送:&nbsp;"爱尚╭ァ ***@***.******@***.***&gt;; 主题:&nbsp;Re: [Lyn-Wu/Lyn] D2NN相关问题 (#2) 经过我的仔细思考，我现在对这个代码完全理解了 ，老哥你的电脑配置如何？我把batchsize改成了10的话1个epoch需要5个小时，如果batchsize是64，计算机直接死机。 另外 代码中有一行D=dL_dL/(N_lmb) 我个人觉得这行多余了，无论是远场计算还是近场计算 都不要这一行 请问方便交流一下么，我对层之间的光传播不是很懂，用的是啥衍射公式呢 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[babypure]

您好 感谢来信 抱歉我不做这个方向了 可以参考相关论文 祝好！ Best wishes, Xuechun

…

________________________________ From: wqh-cyber ***@***.***> Sent: 10 May 2021 14:25 To: Lyn-Wu/Lyn ***@***.***> Cc: Xuechun Wang ***@***.***>; Author ***@***.***> Subject: Re: [Lyn-Wu/Lyn] D2NN相关问题 (#2) 衍射神经网络的远场计算公式和近场计算我印象中有一篇论文中提到过，你那个代码是纯振幅调制而且训练过程不收敛，最好改成纯相位调制。 还有就是可见光的衍射神经网络你有研究么？ 可以加我 QQ: 1205134271&nbsp; &nbsp;

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "Lyn-Wu/Lyn" ***@***.***&gt;; 发送时间:&nbsp;2021年5月1日(星期六) 中午11:14 ***@***.***&gt;; 抄送:&nbsp;"爱尚qァ ***@***.******@***.***&gt;; 主题:&nbsp;Re: [Lyn-Wu/Lyn] D2NN相关问题 (#2) 经过我的仔细思考，我现在对这个代码完全理解了 ，老哥你的电脑配置如何？我把batchsize改成了10的话1个epoch需要5个小时，如果batchsize是64，计算机直接死机。 另外 代码中有一行D=dL_dL/(N_lmb) 我个人觉得这行多余了，无论是远场计算还是近场计算 都不要这一行 请问方便交流一下么，我对层之间的光传播不是很懂，用的是啥衍射公式呢 ― You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. ― You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FLyn-Wu%2FLyn%2Fissues%2F2%23issuecomment-836696275&data=04%7C01%7C%7C43231370be2d4c691b4808d913b7195a%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637562499388009136%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=hUUxRVa2YlZ%2Fbtd2TqiW%2FOsoJz2xm1k%2BjDFOh8b8O7A%3D&reserved=0>, or unsubscribe<https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAMEGEBE7VM3WNKHJ5MGF5JLTM7NFBANCNFSM4H6ZU2EA&data=04%7C01%7C%7C43231370be2d4c691b4808d913b7195a%7C569df091b01340e386eebd9cb9e25814%7C0%7C0%7C637562499388014111%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=QXqu58HkjvWkdkTWJOb6msueWawuJFWrl2DT291QfRg%3D&reserved=0>.