def main(fnModel, fnVolIn, fnMask, sampling, fnVolOut):
if fnModel=="default":
fnModel= XmippScript.getModel("deepRes", "model_w13.h5")
elif fnModel=="highRes":
fnModel= XmippScript.getModel("deepRes", "model_w7.h5")
model = load_model(fnModel)
manager = VolumeManager(fnVolIn, fnMask)
Y = model.predict_generator(manager, manager.getNumberOfBlocks())
if fnModel == XmippScript.getModel("deepRes", "model_w13.h5"):
model = 1
if fnModel == XmippScript.getModel("deepRes", "model_w7.h5"):
model = 2
return produceOutput(fnVolIn, fnMask, model, sampling, Y, fnVolOut)
def generatePerceptualLoss(image_shape):
from xmipp_base import XmippScript
from keras.layers import Input
from keras.models import load_model, Model
import keras.backend as K
effectiveSize = 50000
ignoreCTF = True
modelTypeDir = "keras_models/%sPhaseFlip_Invert/nnetData_%d/tfchkpoints_0" % (
"no" if ignoreCTF else "", effectiveSize)
modelTypeDir = XmippScript.getModel("deepConsensus", modelTypeDir)
modelFname = os.path.join(modelTypeDir, "deepModel.hdf5")
def perceptual_loss(y_true, y_pred):
input_tensor = Input(image_shape[:-1] + (1, ))
evalModel = load_model(modelFname)
targetLayer = 'activation_4'
out = input_tensor
for layer in evalModel.layers[1:]:
# print(layer.name)
out = layer(out)
if layer.name == targetLayer: break
loss_model = Model(inputs=input_tensor, outputs=out)
loss_model.trainable = False
return K.mean(K.square(loss_model(y_true) - loss_model(y_pred)))
return perceptual_loss
def run(self):
params = " -i %s " % self.getParam('-i')
if self.checkParam('-i2'):
params += " -i2 %s " % self.getParam('-i2')
params += " -o %s " % self.getParam('-o')
if self.checkParam('--checkpoint'):
params += " --deepLearningModelPath %s " % os.path.expanduser(
self.getParam("--checkpoint"))
else:
params += " --deepLearningModelPath %s " % XmippScript.getModel(
"deepEMhancer",
"production_checkpoints/deepEMhancer_tightTarget.hd5")
if self.checkParam('--sampling_rate'):
params += " --samplingRate %f" % self.getDoubleParam(
'--sampling_rate')
if self.checkParam('--binaryMask'):
params += " --binaryMask %s " % (os.path.abspath(
self.getParam('--binaryMask')))
elif self.checkParam('--noise_stats_mean'):
params += " --noiseStats %f %f " % (
self.getDoubleParam('--noise_stats_mean'),
self.getDoubleParam('--noise_stats_std'))
if self.checkParam('--cleaningStrengh'):
params += " --cleaningStrengh %f " % self.getDoubleParamWithDefault(
'--cleaningStrengh', defaultVal=-1)
params += "-g %s " % self.getParam("-g")
params += "-b %s " % self.getParam("-b")
cmd = "deepemhancer"
print(cmd + " " + params)
self.runCondaCmd(cmd, params)