Open Ken1256 opened 5 years ago
Hi @Ken1256
Thank you for following our works.
A1: If your aim is image deblurring, you can just train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. A2: Since our works focus on the dual-branch architecture, we only adopted the most commonly used MSE loss, which will generate over-smoothed results... If you want to recovery details and improve the visual results you can introduce the adversary loss during training. Another approach to improve the quality of output is to use a high-quality training dataset. The HR images in GOPRO dataset still have some noises or undesired artifacts due to its limitation of the hardware.
Wish these answers can help you. If you still have any question, be free to contact us.
Hi @Ken1256
Thank you for following our works.
A1: If your aim is image deblurring, you can just train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. A2: Since our works focus on the dual-branch architecture, we only adopted the most commonly used MSE loss, which will generate over-smoothed results... If you want to recovery details and improve the visual results you can introduce the adversary loss during training. Another approach to improve the quality of output is to use a high-quality training dataset. The HR images in GOPRO dataset still have some noises or undesired artifacts due to its limitation of the hardware.
Wish these answers can help you. If you still have any question, be free to contact us.
sorry, where can i find train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. can it deblur 1920x1080 pictures? how to set , thx
A1:The definition of sub-net is in GFN/networks/GFN_4x.py. A2:
On Sun, Jun 13, 2021 at 2:20 PM FrankLinxzx @.***> wrote:
Hi @Ken1256 https://github.com/Ken1256
Thank you for following our works.
A1: If your aim is image deblurring, you can just train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. A2: Since our works focus on the dual-branch architecture, we only adopted the most commonly used MSE loss, which will generate over-smoothed results... If you want to recovery details and improve the visual results you can introduce the adversary loss during training. Another approach to improve the quality of output is to use a high-quality training dataset. The HR images in GOPRO dataset still have some noises or undesired artifacts due to its limitation of the hardware.
Wish these answers can help you. If you still have any question, be free to contact us.
sorry, where can i find train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. can it deblur 1920x1080 pictures? how to set , thx
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jacquelinelala/GFN/issues/6#issuecomment-860160639, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACQXNMJN42IJ7DEX2IQIWXLTSRE2DANCNFSM4GI6EEJA .
A2: This network could test pictures of any size.
On Sun, Jun 13, 2021 at 2:20 PM FrankLinxzx @.***> wrote:
Hi @Ken1256 https://github.com/Ken1256
Thank you for following our works.
A1: If your aim is image deblurring, you can just train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. A2: Since our works focus on the dual-branch architecture, we only adopted the most commonly used MSE loss, which will generate over-smoothed results... If you want to recovery details and improve the visual results you can introduce the adversary loss during training. Another approach to improve the quality of output is to use a high-quality training dataset. The HR images in GOPRO dataset still have some noises or undesired artifacts due to its limitation of the hardware.
Wish these answers can help you. If you still have any question, be free to contact us.
sorry, where can i find train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. can it deblur 1920x1080 pictures? how to set , thx
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jacquelinelala/GFN/issues/6#issuecomment-860160639, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACQXNMJN42IJ7DEX2IQIWXLTSRE2DANCNFSM4GI6EEJA .
A1:The definition of sub-net is in GFN/networks/GFN_4x.py. A2: … On Sun, Jun 13, 2021 at 2:20 PM FrankLinxzx @.***> wrote: Hi @Ken1256 https://github.com/Ken1256 Thank you for following our works. A1: If your aim is image deblurring, you can just train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. A2: Since our works focus on the dual-branch architecture, we only adopted the most commonly used MSE loss, which will generate over-smoothed results... If you want to recovery details and improve the visual results you can introduce the adversary loss during training. Another approach to improve the quality of output is to use a high-quality training dataset. The HR images in GOPRO dataset still have some noises or undesired artifacts due to its limitation of the hardware. Wish these answers can help you. If you still have any question, be free to contact us. sorry, where can i find train the _DeblurringMoudle and deblurring loss without any modification. The deblur_out is the final RGB deblurred output. can it deblur 1920x1080 pictures? how to set , thx — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACQXNMJN42IJ7DEX2IQIWXLTSRE2DANCNFSM4GI6EEJA .
1.this one ? 2.where to define the picture size still cant over 320x176 pixels thx
Q1: Only train _DeblurringMoudle and return deblur_out is right or need modify some code? Q2: Is possible improve image sharpness, anti-aliasing, Denoising and recovery details? THX