Mean and STD values regarding our dataset is required

Nikronic / CoarseNet

Modified version of U-net based on "Convolutional Networks for Biomedical Image Segmentation (Ronneberger et al., 2015)" paper.

https://arxiv.org/abs/1505.04597

MIT License

25 stars 2 forks source link

Mean and STD values regarding our dataset is required #11

Open Nikronic opened 5 years ago

Nikronic commented 5 years ago

We need to calculate mean and std of the whole dataset lazily (online) BEFORE applying halftone.

Actually, The output is something similar to ground truth and by the way we can see in the paper they have mentioned that ground truth is normalized ([-1, 1]). And we cannot calculate mean and std of halftone because of random operations.

Nikronic commented 5 years ago

Please see commit https://github.com/Nikronic/CoarseNet/commit/a1a5300e3eaa5ca037e1385256a91176cf730139. ref: https://discuss.pytorch.org/t/about-normalization-using-pre-trained-vgg16-networks/23560/9

Furthermore, for test purposes on out 9 image subset, here is the mean and std values: mean= [0.3918, 0.3725, 0.3191] std=[0.4881, 0.4835, 0.4661]

Nikronic commented 5 years ago

Please see commit https://github.com/Nikronic/CoarseNet/commit/81a229fb5b8e3753331b9ab98e3d64b6924bd48c .

Nikronic commented 5 years ago

How to deal with halftone if our images are normalized???