READ ME.txt RSGAN

最新推荐文章于 2024-01-20 08:15:00 发布

图像小白Janet

最新推荐文章于 2024-01-20 08:15:00 发布

阅读量218

点赞数

分类专栏： SRGAN 文章标签： pytorch

本文链接：https://blog.csdn.net/qq_39751446/article/details/118144514

版权

SRGAN 专栏收录该内容

3 篇文章 1 订阅

订阅专栏

READ ME.txt

SRGAN1.0 the original code

SRGAN 1.0–>1.1 because the modified test.py and eval.py cannot run the srresnet structure
so from the website:https://blog.csdn.net/qianbin3200896/article/details/104181552
get the accurate code and add test_srresent.py as well as eval_srresent.py to especially run srrssnet stucture

SRGAN 1.1–>1.2 Try to train the module by COCO dataset to make sure the code make sence
first change the ngpu from 2 to 1 to adapt to our environment
A problem come out :torch.nn.modules.module.ModuleAttributeError: ‘Generator’ object has no attribute ‘module’ so change to
the sulotion is to delete the 'moodule ’ part than run the code
the srresnet training code save as checkpoint_srresnet1.pth while the checkpoint_srresnet.pth is the official weight
the srgan training code save as checkpoint_srgan1.pth while the checkpoint_srgan.pth is the cfficial weight
it works well even When test :(epoch == 2 )
srresnet:
the bicubic picture
bicubic result:372*520