Open huang-hz opened 1 year ago
nono,Your test is probably right
I trained in a datatset,which contains 18000 pics including 3 types weather , in 200 epochs and it has converged. And i get the same psnr maybe you can train it by yourself,to check whether something wrong in my train if you improve the psnr,please tell me ( ̄▽ ̄)~*
I've trained a better weight of transweater, you can download it from: https://pan.baidu.com/s/1yBn1GyOTLAx4oS6-kbk6xw?pwd=18o3
score of raindrop is tested on test_a.zip from raindrop dataset
Thank you for your promotion!
I'll mention your work in the readme file
Thanks for both of your contributions! The reason for the incompatibility may be that you tested on different channels. It seems that the PSNR is much higher on Y channel than on RGB channels.
I've trained a better weight of transweater, you can download it from: https://pan.baidu.com/s/1yBn1GyOTLAx4oS6-kbk6xw?pwd=18o3
score of raindrop is tested on test_a.zip from raindrop dataset
refresh the link 链接:https://pan.baidu.com/s/1joD-vNwtavSDLjvFuZtRHg?pwd=1z65 提取码:1z65
Thanks for fixing and sharing! When i test the provided weight file by test_raindropa.py on test_a dataset, i get a val psnr of 28.88. Is here something wrong with my test?