Implementing hypercolumn in neural nets with tensorflow

Question

I was trying to add hypercolumns idea to my neural net model. How can I do this using tensorflow?

Answer 1

Here's a basic example.

Define the layer1,layer2,layer3 etc. then concat them below and now you have your hyper column.

conv1 = tf.nn.conv2d(image, weights, [1, 1, 1, 1], 'SAME')
layer1 = tf.image.resize_bilinear(conv1, (224, 224))
...
layer2 = tf.image.resize_bilinear(conv2, (224, 224))
...
layer3 = tf.image.resize_bilinear(conv3, (224, 224))
return(tf.concat(0, [layer1, layer2,layer3]))