class TensorflowWrapper:
def __init__(self, identifier, inputs, endpoints: dict, session, *args, **kwargs):
import tensorflow as tf
self._inputs = inputs
self._endpoints = endpoints
self._session = session or tf.compat.v1.Session()
self._extractor = ActivationsExtractorHelper(identifier=identifier, get_activations=self.get_activations,
preprocessing=None, *args, **kwargs)
self._extractor.insert_attrs(self)
@property
def identifier(self):
return self._extractor.identifier
@identifier.setter
def identifier(self, value):
self._extractor.identifier = value
def __call__(self, *args, **kwargs): # cannot assign __call__ as attribute due to Python convention
return self._extractor(*args, **kwargs)
def get_activations(self, images, layer_names):
layer_tensors = OrderedDict((layer, self._endpoints[
layer if (layer != 'logits' or layer in self._endpoints) else next(reversed(self._endpoints))])
for layer in layer_names)
layer_outputs = self._session.run(layer_tensors, feed_dict={self._inputs: images})
return layer_outputs
def graph(self):
import networkx as nx
g = nx.DiGraph()
for name, layer in self._endpoints.items():
g.add_node(name, object=layer, type=type(layer))
g.add_node("logits", object=self.logits, type=type(self.logits))
return g
class KerasWrapper:
def __init__(self, model, preprocessing, identifier=None, *args, **kwargs):
"""
:param model: a keras model with a function `preprocess_input`
that will later be called on the loaded numpy image
"""
self._model = model
identifier = identifier or model.name
self._extractor = ActivationsExtractorHelper(
identifier=identifier,
get_activations=self.get_activations,
preprocessing=preprocessing,
*args,
**kwargs)
self._extractor.insert_attrs(self)
@property
def identifier(self):
return self._extractor.identifier
@identifier.setter
def identifier(self, value):
self._extractor.identifier = value
def __call__(
self, *args, **kwargs
): # cannot assign __call__ as attribute due to Python convention
return self._extractor(*args, **kwargs)
def get_activations(self, images, layer_names):
from keras import backend as K
input_tensor = self._model.input
layers = [
layer for layer in self._model.layers if layer.name in layer_names
]
layers = sorted(layers,
key=lambda layer: layer_names.index(layer.name))
if 'logits' in layer_names:
layers.insert(layer_names.index('logits'), self._model.layers[-1])
assert len(layers) == len(layer_names)
layer_out_tensors = [layer.output for layer in layers]
functor = K.function([input_tensor] + [K.learning_phase()],
layer_out_tensors) # evaluate all tensors at once
layer_outputs = functor([images, 0.]) # 0 to signal testing phase
return OrderedDict([
(layer_name, layer_output)
for layer_name, layer_output in zip(layer_names, layer_outputs)
])
def __repr__(self):
return repr(self._model)
def graph(self):
import networkx as nx
g = nx.DiGraph()
for layer in self._model.layers:
g.add_node(layer.name, object=layer, type=type(layer))
for outbound_node in layer._outbound_nodes:
g.add_edge(layer.name, outbound_node.outbound_layer.name)
return g
class PixelModel:
def __init__(self):
self._extractor = ActivationsExtractorHelper(
identifier='pixels',
preprocessing=None,
get_activations=self._pixels_from_paths)
self._extractor.insert_attrs(self)
@property
def identifier(self):
return self._extractor.identifier
@identifier.setter
def identifier(self, value):
self._extractor.identifier = value
def __call__(
self, *args, **kwargs
): # cannot assign __call__ as attribute due to Python convention
return self._extractor(*args, **kwargs)
def _pixels_from_paths(self, paths, layer_names):
np.testing.assert_array_equal(layer_names, ['pixels'])
pixels = [self._parse_image(path) for path in paths]
return OrderedDict([('pixels', np.array(pixels))])
def _parse_image(self, path):
image = Image.open(path)
image = image.convert(
'RGB') # make sure everything is in RGB and not grayscale L
image = image.resize((256, 256)) # resize all images to same size
return np.array(image)