def get_layers_and_variables_from_model(model: tf.keras.Model, scope_name=None):
layer_dict = {}
if scope_name is not None:
base_name = scope_name
else:
base_name = model.name
# get Layers
for layer in model.layers:
if isinstance(layer, tf.keras.Model):
sub_model_layer_dict = get_layers_and_variables_from_model(
layer, "{}/{}".format(base_name, layer.name)
)
layer_dict.update(sub_model_layer_dict)
elif isinstance(layer, tf.keras.layers.Layer):
layer_dict["{}/{}".format(base_name, layer.name)] = layer
# get Variables
for attr_name, attr_value in model.__dict__.items():
# NOTE: _train_counter, _test_counter, _predict_counter are
# built-in variables of tf.keras.Model
if attr_name not in [
"_train_counter",
"_test_counter",
"_predict_counter",
] and isinstance(attr_value, tf.Variable):
layer_dict["{}/{}".format(base_name, attr_value.name)] = attr_value
return layer_dict
递归：因为keras.Model中可能包含另一个keras.Model
版本：tf2.5.1
标签：layer,name,get,keras,dict,tf,model From： https://www.cnblogs.com/deepllz/p/16601117.html