How to normalize the image for custom dataset about densenet-tensorflow HOT 3 CLOSED

Hi Join,

All the training images will be used during training.
The number you saw 7440 is related to the random shuffle function of the training set function provided by tensorflow.

With regards!

from densenet-tensorflow.

1. It's ok if there's *.png files in your dataset.
2. You can find normalization function in tensorflow, I remember the name is per_image_standardization or something. Besides, whether to do the input normalization is totally up to you, it's not a must for nowadays neural network.

With regards!

from densenet-tensorflow.

Hi,

I am new to tensorflow. I have a confusion regarding these equations.
image_raw = tf.image.decode_jpeg(image_encoded, channels=3)
image_raw = tf.cast(image_raw , tf.float32) * (1. / 255)

These equations seem to be correct to me for an image. But for a batch of images, how to modify these equations?

I read in the documentation of tensorflow for the function : per_image_standardization. It linearly scales image to have zero mean and unit norm. But I want the image to be in the range of [0, 1] instead of [0-255]. So, cannot use this function.

from densenet-tensorflow.