search

sanghyun-son / EDSR-PyTorch

PyTorch version of the paper 'Enhanced Deep Residual Networks for Single Image Super-Resolution' (CVPRW 2017)
MIT License
2.42k stars 668 forks source link

Input data normalization #282

Open smiler96 opened 3 years ago

smiler96 commented 3 years ago

i see that the MeanShift has been applied to input data as the zero normalization, while the data' values is not in the -1 ~ 1 range. Could this effect the performance? have the input image data just been divided by 255.0 (data value in 0 ~ 1 range) as the normalization to train the model and how is the final performance? thanks for your reply!

HolmesShuan commented 3 years ago

have the input image data just been divided by 255.0 (data value in 0 ~ 1 range) as the normalization to train the model and how is the final performance?

FYI. def np2Tensor(*args, rgb_range=255):.