luoru / correlation1d-tf

DispNet correlation layer tensorflow implementation. the code has been tested success!

Correlation1d Tensorflow version

DispNet correlation1d layer tensorflow implementation. the code has been tested success! This code mainly based on (https://github.com/sampepose/flownet2-tf.git), which has implement the correlation of flownet. Other version of correlation1d in github still have some bugs

Compiling

just run compille.sh

correlation1d example

$ python

>>> import tensorflow as tf
>>> x1 = tf.random_normal([1, 10, 10, 4])
>>> x2 = tf.random_normal([1, 10, 10, 4])
>>> conv_x1 = tf.layers.conv2d(x1, 10, 3, 1, 'same', name='conv_x1')
>>> conv_x2 = tf.layers.conv2d(x2, 10, 3, 1, 'same', name='conv_x2')
>>> corr = correlation1d(conv_x1, conv_x1, 1, 4, 1, 1, 4)
>>> grad_corr2x1 = tf.gradients(corr, x1)