About training result - Githubissues

HW-VMCL commented 2 years ago

Hi, I trained the BMNet on RTX 3090 with ISTD dataset, but PSNR on valid set converge to 28.699. I can only train the net with batch_size=2 not 4 as it reported in the paper. Moreover, Could you please provide a pretrained model? Thanks!

Shyla1999 commented 1 week ago

Hi, I trained the BMNet on RTX 3090 with ISTD dataset, but PSNR on valid set converge to 28.699. I can only train the net with batch_size=2 not 4 as it reported in the paper. Moreover, Could you please provide a pretrained model? Thanks!

你好，为什么你可以test出来这么高的PSNR啊，我test的时候只有3.3左右，救命

HW-VMCL commented 1 week ago

Hi, I trained the BMNet on RTX 3090 with ISTD dataset, but PSNR on valid set converge to 28.699. I can only train the net with batch_size=2 not 4 as it reported in the paper. Moreover, Could you please provide a pretrained model? Thanks!

你好，为什么你可以test出来这么高的PSNR啊，我test的时候只有3.3左右，救命

I'm not sure, you may email the authors for an answer. Or maybe you can check the PSNR calculation function.

Shyla1999 commented 1 week ago

您好，我没有找到colortrans那部分模型的预训练权重，不知道是不是作者说的那几个预训练权重，所以我只在mainnet部分做了预训练权重的添加，然后做了测试，但是出来确实是结果很低，想问一下您在colortrans部分假的预训练权重是在哪里呀，非常感谢！

---- Replied Message ---- | From | @.> | | Date | 11/04/2024 19:41 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [KevinJ-Huang/BMNet] About training result (Issue #2) |

Hi, I trained the BMNet on RTX 3090 with ISTD dataset, but PSNR on valid set converge to 28.699. I can only train the net with batch_size=2 not 4 as it reported in the paper. Moreover, Could you please provide a pretrained model? Thanks!

你好，为什么你可以test出来这么高的PSNR啊，我test的时候只有3.3左右，救命

I'm not sure, you may email the authors for an answer. Or maybe you can check the PSNR calculation function.

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

KevinJ-Huang / BMNet

About training result #2