search

WeishengDong / DPDNN_PyTorch

23 stars 6 forks source link

recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) #1

Open wangweiwei104 opened 4 years ago

wangweiwei104 commented 4 years ago

Hi, According to the formula of iteration

Then,Let

Therefore, the code(Line 178 in model.py) recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) should be recon = torch.mul((1 - delta - eta), x) + torch.mul(eta, v) + torch.mul(delta, y)

Is it right?

WeishengDong commented 4 years ago

My former student, i.e., the second author of our paper wrote the code. Then, I suggest you write email to ask her.

Thanks for your interests.

Weisheng

发件人: bounces+848413-485e-wsdong=mail.xidian.edu.cn@sgmail.github.com bounces+848413-485e-wsdong=mail.xidian.edu.cn@sgmail.github.com 代表 wangweiwei104 发送时间: 2020年7月31日 17:11 收件人: WeishengDong/DPDNN_PyTorch DPDNN_PyTorch@noreply.github.com 抄送: Subscribed subscribed@noreply.github.com 主题: [WeishengDong/DPDNN_PyTorch] recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) (#1)

Hi, According to the formula of iteration

Then,Let https://camo.githubusercontent.com/a7d81e9d196de3bba90afa430664413150f68b83/68747470733a2f2f6c617465782e636f6465636f67732e636f6d2f6769662e6c617465783f253543647069253742343030253744413d492c25354364656c74612532302535436574613d253543657461 https://camo.githubusercontent.com/e7f888e905a57b01dbd97f03fb855a8b056c33ea/68747470733a2f2f6c617465782e636f6465636f67732e636f6d2f6769662e6c617465783f253543647069253742343030253744782535452537422535436c65667425323028253230742b312532302535437269676874253230292537443d2535436c65667425323028253230312d25354364656c74612532302d253543657461253230253230253543726967687425323029782b25354364656c74612532304125354525374254253744792b25354365746125323076

Therefore, the code(Line 178 in model.py) recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) should be recon = torch.mul((1 - delta - eta), x) + torch.mul(eta, v) + torch.mul(delta, y)

Is it right?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/WeishengDong/DPDNN_PyTorch/issues/1 , or unsubscribe https://github.com/notifications/unsubscribe-auth/AKZ3L3PX3FZIOJPPOCYNBZTR6KDAXANCNFSM4PP227GQ . https://github.com/notifications/beacon/AKZ3L3LQLZ7XSLYBKNRY6DDR6KDAXA5CNFSM4PP227G2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4J7JH6HQ.gif

wangweiwei104 commented 4 years ago

My former student, i.e., the second author of our paper wrote the code. Then, I suggest you write email to ask her. Thanks for your interests. Weisheng 发件人: bounces+848413-485e-wsdong=mail.xidian.edu.cn@sgmail.github.com bounces+848413-485e-wsdong=mail.xidian.edu.cn@sgmail.github.com 代表 wangweiwei104 发送时间: 2020年7月31日 17:11 收件人: WeishengDong/DPDNN_PyTorch DPDNN_PyTorch@noreply.github.com 抄送: Subscribed subscribed@noreply.github.com 主题: [WeishengDong/DPDNN_PyTorch] recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) (#1) Hi, According to the formula of iteration Then,Let https://camo.githubusercontent.com/a7d81e9d196de3bba90afa430664413150f68b83/68747470733a2f2f6c617465782e636f6465636f67732e636f6d2f6769662e6c617465783f253543647069253742343030253744413d492c25354364656c74612532302535436574613d253543657461 https://camo.githubusercontent.com/e7f888e905a57b01dbd97f03fb855a8b056c33ea/68747470733a2f2f6c617465782e636f6465636f67732e636f6d2f6769662e6c617465783f253543647069253742343030253744782535452537422535436c65667425323028253230742b312532302535437269676874253230292537443d2535436c65667425323028253230312d25354364656c74612532302d253543657461253230253230253543726967687425323029782b25354364656c74612532304125354525374254253744792b25354365746125323076 Therefore, the code(Line 178 in model.py) recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) should be recon = torch.mul((1 - delta - eta), x) + torch.mul(eta, v) + torch.mul(delta, y) Is it right? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1> , or unsubscribe https://github.com/notifications/unsubscribe-auth/AKZ3L3PX3FZIOJPPOCYNBZTR6KDAXANCNFSM4PP227GQ . https://github.com/notifications/beacon/AKZ3L3LQLZ7XSLYBKNRY6DDR6KDAXA5CNFSM4PP227G2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4J7JH6HQ.gif

Thanks for your reply. Can you provide me with his contact information (e.g. email)?

Swocky commented 3 years ago

My former student, i.e., the second author of our paper wrote the code. Then, I suggest you write email to ask her. Thanks for your interests. Weisheng 发件人: bounces+848413-485e-wsdong=mail.xidian.edu.cn@sgmail.github.com bounces+848413-485e-wsdong=mail.xidian.edu.cn@sgmail.github.com 代表 wangweiwei104 发送时间: 2020年7月31日 17:11 收件人: WeishengDong/DPDNN_PyTorch DPDNN_PyTorch@noreply.github.com 抄送: Subscribed subscribed@noreply.github.com 主题: [WeishengDong/DPDNN_PyTorch] recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) (#1) Hi, According to the formula of iteration Then,Let https://camo.githubusercontent.com/a7d81e9d196de3bba90afa430664413150f68b83/68747470733a2f2f6c617465782e636f6465636f67732e636f6d2f6769662e6c617465783f253543647069253742343030253744413d492c25354364656c74612532302535436574613d253543657461 https://camo.githubusercontent.com/e7f888e905a57b01dbd97f03fb855a8b056c33ea/68747470733a2f2f6c617465782e636f6465636f67732e636f6d2f6769662e6c617465783f253543647069253742343030253744782535452537422535436c65667425323028253230742b312532302535437269676874253230292537443d2535436c65667425323028253230312d25354364656c74612532302d253543657461253230253230253543726967687425323029782b25354364656c74612532304125354525374254253744792b25354365746125323076 Therefore, the code(Line 178 in model.py) recon = torch.mul((1 - delta - eta), v) + torch.mul(eta, x) + torch.mul(delta, y) should be recon = torch.mul((1 - delta - eta), x) + torch.mul(eta, v) + torch.mul(delta, y) Is it right? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1> , or unsubscribe https://github.com/notifications/unsubscribe-auth/AKZ3L3PX3FZIOJPPOCYNBZTR6KDAXANCNFSM4PP227GQ . https://github.com/notifications/beacon/AKZ3L3LQLZ7XSLYBKNRY6DDR6KDAXA5CNFSM4PP227G2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4J7JH6HQ.gif

董老师您好，按照推导确实是这样，但是实际上可以理解为x，y，v的trade off，y系数是delta，x和v的系数和是1-delta，而eta与y无关，所以两者交换系数应该对最终优化结果没有影响。只是感觉代码这么写会有一定误导性，不太清楚出于什么考虑与论文选择不一致。另外，在deblur的时候A就不再是identity matrix，但实际代码中的重构公式与去噪是一致的，而dataset中在取训练数据时最后加入一定的噪声，这里我不太理解为何这样设计，希望您能给一些解释，感谢。