search

swz30 / MPRNet

[CVPR 2021] Multi-Stage Progressive Image Restoration. SOTA results for Image deblurring, deraining, and denoising.
Other
1.17k stars 190 forks source link

model size #79

Closed chenloveheimei closed 3 years ago

chenloveheimei commented 3 years ago

hello,my name is Long. I downloaded your pretrain model and the model's size is about 80M. When I train a new model,I found the model's size is 241M.what should I do can reduce the size? thank you.

adityac8 commented 3 years ago

The discrepancy might occur if you use the arguments of the Deblurring task for Deraining as the Deblurring model is heavier.

dingyan1478 commented 2 years ago

hello,my name is Long. I downloaded your pretrain model and the model's size is about 80M. When I train a new model,I found the model's size is 241M.what should I do can reduce the size? thank you.

您好，我也遇到了这个问题， 我下载的作者提供的去模糊的预训练模型大概是80M左右，可是如果如果我自己训练去雾的模型，得到的模型大概240M左右，图一是预训练的模型，图二是我自己训练的模型。 image image 如果您解决了这个问题，非常感谢您能够帮助我一下，这个问题困扰了我好久。 再次感谢！

hust-lidelong commented 2 years ago

已收到邮件，谢谢

dingyan1478 commented 2 years ago

已收到邮件，谢谢

您好！麻烦请问一下，这个问题您解决了吗？

dingyan1478 commented 2 years ago

已收到邮件，谢谢

很抱歉再次打扰到您，想问一下您的这个问题解决了吗？当我自己训练网络模型的时候会得到大概240M左右的模型，但是加载作者使用的预训练模型的话74M左右，而且，如果在作者预训练的权重上继续进行训练的话，模型的大小还是240左右，是哪里出了问题吗？ 非常期待您的回复！

ak-karimzai commented 6 months ago

@dingyan1478 can you share plz loss dependency and psnr and ssim value in each epoch?