Tensorflow-MultiGPU-VAE-GAN icon indicating copy to clipboard operation
Tensorflow-MultiGPU-VAE-GAN copied to clipboard
verified publisher icon timsainb

Normalizing the images doesn't work

Open alexrakowski opened this issue 8 years ago • 2 comments

Normalizing the dataset to values in range [0,1], done by this line: faces = (faces/255.) doesn't seem to work properly, at least not with the visualizations. Is it meant to somehow improve the training process?

alexrakowski avatar May 18 '17 10:05 alexrakowski

Hi Alex, Normalization/conversion of inputs to float is necessary before you feed the data into the model at least at some point. The model's output are squashed with a sigmoid between 0-1. What about normalization is not working properly?

timsainb avatar May 18 '17 18:05 timsainb

Visualizing the images with matplotlib - after normalization I would get fully black images (which makes kind of sense).

alexrakowski avatar May 18 '17 21:05 alexrakowski