This program automatically computes the receptive field size of the top-most layer of the type Conv2D or PoolMax (the narrowest layer of a FCN or AutoEncoder). It is important that the receptive field size is big enough to capture the characteristic structure of the underlying image.
Usage:
rfsize.py --model MODEL
# MODEL is a previously saved model snapshot while training.
# there should be MODEL.meta, MODEL.index and MODEL.data_...