search

keeper121 / liteflownet-tf2

LiteFlowNet inference realization with TensorFlow 2
17 stars 7 forks source link

why do we need `module_feat()` in `matching()`? #5

Open zhuokaizhao opened 4 years ago

zhuokaizhao commented 4 years ago

I am wondering why would we need module_feat() within matching()? More specifically I did not find any reference from the original paper that says we need to apply a convolutional layer specifically for the level-2 features. Did I miss something here? Thanks!

def module_feat():
  if int_level == 2:
    return tf.keras.layers.Conv2D(filters=64, kernel_size=1, activation=lrelu, padding='valid')
  else:
    return tf.keras.Sequential()