settings.py

import os
from inspect import getsourcefile


# String descriptions of the architectures in use
ARCHITECTURE = "deeplab-xception"
DAMAGE_ARCHITECTURE = 'deeplab-xception'

# Train/validation split proportion
SPLITFACTOR = 0.9
# Minibatch size
BATCH_SIZE = 1

#           0        1             2                3              4              5
CLASSES = [None, "no-damage", "minor-damage", 'major-damage', 'destroyed', 'un-classified']
N_CLASSES = len(CLASSES)

# Classify damage
DAMAGE = True 

# weight file names
MODELSTRING = "%s-%d.hdf5" % (ARCHITECTURE, N_CLASSES)
DMG_MODELSTRING = "damage-%s.hdf5" % DAMAGE_ARCHITECTURE

MODELSTRING_BEST = MODELSTRING.replace(".hdf5", "-best.hdf5")
DMG_MODELSTRING_BEST = DMG_MODELSTRING.replace(".hdf5", "-best.hdf5")

# The maximum length of building patches for individual damage classification
# (unused by default)
DAMAGE_MAX_X = 224
DAMAGE_MAX_Y = 224

# Shape of the training input images
INPUTSHAPE = [1024,1024,6]
# Shape of the target masks
MASKSHAPE = [1024,1024,N_CLASSES]
# The shape of the raw on-disk input images
SAMPLESHAPE = [1024,1024,3]
TARGETSHAPE = INPUTSHAPE

# size of combination pre+post building images (unused by default)
DMG_SAMPLESHAPE = [128,128,3]
DMG_INPUTSHAPE = DMG_SAMPLESHAPE

# The training and validation set files (pickled versions for faster loading, unused by default)
PICKLED_TRAINSET = "trainingsamples.pickle"
PICKLED_VALIDSET = "validationsamples.pickle"

# Base directory of the Python files
HOMEDIR = os.path.abspath(os.path.dirname(getsourcefile(lambda:0)))
# Directories where the training images are located
IMAGEDIRS = ["train/images", "tier3/images"]
# Directories where the .json files describing the images are
LABELDIRS = ["train/labels", "tier3/labels"]
# Directories where the test images are kept
TESTDIRS = ["/data/xview2/test/images"]
# The following is not to be changed: used only to locate the .png file for reading
ALLIMAGEDIRS = IMAGEDIRS + TESTDIRS