Combined database normalisation
codingzebra33 opened this issue · comments
codingzebra33 commented
How was the combined database normalisation in the script group_generator.py (starting from 64 line) done?
Junyong You commented
Hi, I just calculated the mean and std of the images from the combination of the two datasets, which are hard coded in line 64-69. But actually if you are using ImageNet pretrained weights (I would strongly suggest so), you should do ImageNet preprocess (like lines 60, 61 in TensorFlow or the mean and std in Pytorch).