def get_feature_maps(self, layer_names, img_paths):
"""
Return the feature maps outputted by specified layers during prediction of specified images.
:param layer_names: a list of the name of the layer whose output you want to capture
:param img_paths: a list of paths to the jpg or png file
:return: refer to the description of param 'output_style'
"""
util.enforce_list(layer_names, img_paths)
model_with_output_layers = self.get_model_with_output_layers(layer_names)
img_tensors = util.from_img_paths(img_paths, target_size=self.input_shape[1:-1], preprocess=self.preprocess_style)
feature_maps = model_with_output_layers.predict(img_tensors)
return util.to_indexable(util.enforce_list(feature_maps))
def wrapper(*args):
outputs = []
img_paths = args[1]
img_paths = util.enforce_list(img_paths)
for img_path in img_paths:
outputs.append(func(args[0], img_path))
return outputs[0] if (len(outputs) == 1) else outputs
def get_outputs_of_layers(self, layer_names):
"""
Return a list of symbolic outputs of specified convolution layers.
Helper method of `self.get_model_with_output_layers`.
:param layer_names: a list of layer names
:return: a list of symbolic outputs of selected convolution layers
"""
# convert one-item `layer_names` into a list of that one item
layer_names = util.enforce_list(layer_names)
# append symbolic output tensors of convolution layers whose names are in `layer_names`
outputs = []
for layer_name in layer_names:
layer = self.model.get_layer(layer_name) # captures 'layer not found' errors
if isinstance(layer, keras.layers.Conv2D):
outputs.append(layer.output)
else:
raise ValueError('Layer with name "{}" is not a 2D convolution layer.'.format(layer_name))
return outputs