Zero-DCE icon indicating copy to clipboard operation
Zero-DCE copied to clipboard

参数请教?

Open hejing-maker opened this issue 5 years ago • 2 comments

首先感谢你的开源,我在读代码的时候发现损失函数的权重参数和论文中有些不同,论文中提到:The weights Wcol and WtvA are set to 0.5, and 20,但是我看代码中设置为: Loss_TV = 200*L_TV(A)

		loss_spa = torch.mean(L_spa(enhanced_image, img_lowlight))

		loss_col = 5*torch.mean(L_color(enhanced_image))

		loss_exp = 10*torch.mean(L_exp(enhanced_image))

请问影响大吗?

hejing-maker avatar Jan 08 '21 06:01 hejing-maker

请以代码为准,因为论文里面是没有考虑troch.mean的,代码里面是最优的。谢谢!

hejing-maker [email protected] 于2021年1月8日周五 下午2:57写道:

首先感谢你的开源,我在读代码的时候发现损失函数的权重参数和论文中有些不同,论文中提到:The weights Wcol and WtvA are set to 0.5, and 20,但是我看代码中设置为: Loss_TV = 200*L_TV(A)

  loss_spa = torch.mean(L_spa(enhanced_image, img_lowlight))



  loss_col = 5*torch.mean(L_color(enhanced_image))



  loss_exp = 10*torch.mean(L_exp(enhanced_image))

请问影响大吗?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Li-Chongyi/Zero-DCE/issues/19, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIEVQENRBO3AAHYPEHUBGKDSY2UGRANCNFSM4V2BGEPQ .

Li-Chongyi avatar Jan 08 '21 14:01 Li-Chongyi

你好,请问你在使用他的数据集进行训练的时候,有碰到损失函数不收敛的问题吗?

DuXikb avatar Aug 08 '23 07:08 DuXikb