How to perform the (reversed) gamma correction

[sczhou]

Hi Xuanchi,

I notice that you performed reversed gamma correction before convolution with blur kernel. And you demonstrated it works a lot. I wonder how did you perform the reversed gamma correction and gamma correction. Because it may be hard to get a general inverse camera response function for different cameras.

Thanks!

[xrenaa]

Thanks for your question and sorry for late reply due to the exam weeks.
The code for making blur is as below:

_im = tf.sign(im) * (tf.abs(im)) ** (2.2)
B,H,W,C = im.get_shape()
c1 = tf.nn.conv2d(_im[:,:,:,0:1], kernel, strides=[1,1,1,1], padding='SAME')
c2 = tf.nn.conv2d(_im[:,:,:,1:2], kernel, strides=[1,1,1,1], padding='SAME')
c3 = tf.nn.conv2d(_im[:,:,:,2:3], kernel, strides=[1,1,1,1], padding='SAME')
result = tf.concat([c1,c2,c3], axis = 3)
result = tf.sign(result) * (tf.abs(result)) ** (1/2.2)

This works for specific scenarios. We set a commonly used parameter.

[sczhou]

Got it! Thanks for your answer!