CODEWITHSUNDEEP
machine-learningtensorflowcomputer-vision
j35t3r
j35t3r

Reputation: 1533

How to calculate unit variance in tensorflow?

My in put is a set of images and I want to calculate the univariance over my images. But when try to check with numpy, the unit variance should give 1 in the end. What I am doing wrong in my code?

def pre_processing(img_list, zero_mean=True, unit_var=True):
    with tf.device('/cpu:0'):
        tn_img0 = img_list[0][1]
        tn_img1 = img_list[1][1]

        t_img = tn_img0
        # t_img = tf.concat([tn_img0, tn_img1], axis=0)
        rgb_mean, rgb_var = tf.nn.moments(t_img, [0, 1])

        if zero_mean:
            tn_img0 = tf.subtract(img_list[0][1], rgb_mean)
            tn_img1 = tf.subtract(img_list[1][1], rgb_mean)

        if unit_var:
            tn_img0 = tf.divide(tn_img0, rgb_var)
            tn_img1 = tf.divide(tn_img1, rgb_var)

Upvotes: 0

Views: 309

Answers (1)

Vijay Mariappan
Vijay Mariappan

Reputation: 17201

You should divide by the standard deviation to get a unit variance of your inputs. So change your code to:

tn_img0 = tf.divide(tn_img0, tf.sqrt(rgb_var))
tn_img1 = tf.divide(tn_img1, tf.sqrt(rgb_var))

Upvotes: 1

Related Questions