huawei-noah / Pretrained-IPT

Apache License 2.0
434 stars 63 forks source link

超分辨率模型如何训练灰度图 #51

Open dongyan007 opened 1 year ago

dongyan007 commented 1 year ago

我将灰度图叠加三份作成rgb图,效果有彩色像素,如下 image

请问如何训练灰度图

HantingChen commented 1 year ago

你好,可以改变输入参数num_channels为1,使网络的输入和输出channel数都为1,然后重新训练网络

dongyan007 commented 1 year ago

老师,您好:

我的训练数据是单通道灰度图 我做了如下的操作: 1.修改--n_colors

2.修改srdata的getitem方法

3.取消了sub_mean和add_mean

训练完成后,我对下图进行测试:

,结果却是这样的

,请问这是怎么回事呢

董琰 @.***

 

------------------ 原始邮件 ------------------ 发件人: "huawei-noah/Pretrained-IPT" @.>; 发送时间: 2022年11月14日(星期一) 下午2:54 @.>; @.**@.>; 主题: Re: [huawei-noah/Pretrained-IPT] 超分辨率模型如何训练灰度图 (Issue #51)

你好,可以改变输入参数num_channels为1,使网络的输入和输出channel数都为1,然后重新训练网络

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>