def dsen2_20(d10, d20):
# Input to the funcion must be of shape:
# d10: [x,y,4] (B2, B3, B4, B8)
# d20: [x/2,y/4,6] (B5, B6, B7, B8a, B11, B12)
# deep: specifies whether to use VDSen2 (True), or DSen2 (False)
border = 8
p10, p20 = get_test_patches(d10, d20, patchSize=128, border=border)
p10 /= SCALE
p20 /= SCALE
test = [p10, p20]
prediction = _predict(test, model_filename=MDL_PATH_20m_AESR)
images = recompose_images(prediction, border=border, size=d10.shape)
images *= SCALE
return images
def DSen2_20(d10, d20, deep=False):
# Input to the funcion must be of shape:
# d10: [x,y,4] (B2, B3, B4, B8)
# d20: [x/2,y/4,6] (B5, B6, B7, B8a, B11, B12)
# deep: specifies whether to use VDSen2 (True), or DSen2 (False)
border = 8
p10, p20 = get_test_patches(d10, d20, patchSize=128, border=border)
p10 /= SCALE
p20 /= SCALE
test = [p10, p20]
input_shape = ((4, None, None), (6, None, None))
prediction = _predict(test, input_shape, deep=deep)
images = recompose_images(prediction, border=border, size=d10.shape)
images *= SCALE
return images
fileList = [
os.path.basename(x)
for x in sorted(glob.glob(path + folder + "*SAFE"))
]
for dset in fileList:
start = time.time()
print("Timer started.")
print("Predicting: {}.".format(dset))
train, image_size = OpenDataFilesTest(path + folder + dset,
args.run_60, SCALE,
args.true)
prediction = model.predict(train, batch_size=8, verbose=1)
prediction_file = model_nr + "-predict"
# np.save(path + 'test/' + dset + '/' + prediction_file + 'pat', prediction * SCALE)
images = recompose_images(prediction,
border=border,
size=image_size)
print("Writing to file...")
np.save(path + folder + dset + "/" + prediction_file,
images * SCALE)
end = time.time()
print("Elapsed time: {}.".format(end - start))
sys.exit(0)
if args.resume_file:
print("Will resume from the weights {}".format(args.resume_file))
model.load_weights(args.resume_file)
model_nr = args.resume_file[-20:-13]
print("Changing the model number to: {}".format(model_nr))
else: