setting of pre-train model

[hhcs9527]

Hi Thanks for the great work! I am wondering what is the setting of the pre-trained model that you provided for deblurring? Is it n_feat=40, scale_unetfeats=40, scale_orsnetfeats=32 ? Or n_feat=96, scale_unetfeats=48, scale_orsnetfeats=32 ?

Do you have them both? Are they suitable for PyTroch 1.8?

Thanks!

[adityac8]

Hi @hhcs9527

The pre-tained model provided at https://drive.google.com/file/d/1QwQUVbk6YVOJViCsOKYNykCsdJSVGRtb/view?usp=sharing uses the settings at https://github.com/swz30/MPRNet/blob/1b8ec5c94506b7abfc0d71a62b58d32aa867bcfb/Deblurring/MPRNet.py#L239

We have only tested our models at PyTorch 1.1 If you are able to test it on other versions, please let us know.

Thanks

[hhcs9527]

Thanks for your information. Could you provide the ssim under the setting of n_feat=40, scale_unetfeats=40, scale_orsnetfeats=32? Thanks

[adityac8]

Hi @hhcs9527 You can find the details about our light-weight Deblurring model at #38 Once you have the images, you can run evaluate_GOPRO_HIDE.m

Thanks

[hhcs9527]

Thanks for your information, I tested both of them on RTX3090 with pytorch 1.8 and it works fine! For the record,

Method. | PSNR. | SSIM origin | 32.658 | 0.959 lite version| 31.87 | 0.952

Thanks again !!!

[adityac8]

Thank you for letting us know that our model is able to get the same performance with newer pytorch versions.

[dingyan1478]

Hi @hhcs9527

The pre-tained model provided at https://drive.google.com/file/d/1QwQUVbk6YVOJViCsOKYNykCsdJSVGRtb/view?usp=sharing uses the settings at

https://github.com/swz30/MPRNet/blob/1b8ec5c94506b7abfc0d71a62b58d32aa867bcfb/Deblurring/MPRNet.py#L239

We have only tested our models at PyTorch 1.1 If you are able to test it on other versions, please let us know.

Thanks HI @adityac8 您好，按照您代码中默认的参数n_feat=96, scale_unetfeats=48, scale_orsnetfeats=32设置，您提供的预训练模型：https://drive.google.com/file/d/1QwQUVbk6YVOJViCsOKYNykCsdJSVGRtb/view?usp=sharing，模型的大小为80M左右，可是我 按照您代码中默认的参数n_feat=96, scale_unetfeats=48, scale_orsnetfeats=32训练的模型大小在240M左右，为什么会这样呢？是我的模型哪里除了问题吗？

[dingyan1478]

Hi Thanks for the great work! I am wondering what is the setting of the pre-trained model that you provided for deblurring? Is it n_feat=40, scale_unetfeats=40, scale_orsnetfeats=32 ? Or n_feat=96, scale_unetfeats=48, scale_orsnetfeats=32 ?

Do you have them both? Are they suitable for PyTroch 1.8?

Thanks!

Sorry to bother you!

I would like to ask you whether you use the pre-trained model provided by the author for testing, or do you train yourself? According to the parameter settings mentioned by the author, the size of the model I trained is about 270M, but the size provided by the author is about 80M. Have you encountered this problem?

def init(self, in_c=3, out_c=3, n_feat=96, scale_unetfeats=48, scale_orsnetfeats=32, num_cab=8, kernel_size=3, reduction=4, bias=False):