Some deep learning task, such as SSD and neural-style may need to extract previous lay

<a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="/us

How to get layer (output) by name? about tensorlayer HOT 4 CLOSED

tensorlayer commented on May 13, 2024

How to get layer (output) by name?

from tensorlayer.

Comments (4)

zsdonghao commented on May 13, 2024 1

@JindongJiang Good idea, indeed.

I think we can create 2 functions for Layer class : network.get_layers_by_name("conv2") and network.get_params_by_name("W").

network.get_params_by_name("W") should be similar with tl.layer. get_variables_with_name() as follow:

def get_variables_with_name(name, train_only=True, printable=False):
    """Get variable list by a given name scope.

    Examples
    ---------
    >>> dense_vars = get_variable_with_name('dense', True, True)
    """
    print("  Get variables with %s" % name)
    t_vars = tf.trainable_variables() if train_only else tf.all_variables()
    d_vars = [var for var in t_vars if name in var.name]
    if printable:
        for idx, v in enumerate(d_vars):
            print("  got {:3}: {:15}   {}".format(idx, v.name, str(v.get_shape())))
    return d_vars

You can try this code by adding it to Layer:

def get_params_by_name(self, name, printable=False):
    print("  Get params in a network with %s" % name)
    t_vars = self.all_params
    d_vars = [var for var in t_vars if name in var.name]
    if printable:
        for idx, v in enumerate(d_vars):
            print("  got {:3}: {:15}   {}".format(idx, v.name, str(v.get_shape())))
    return d_vars

from tensorlayer.

JindongJiang commented on May 13, 2024

@zsdonghao Thx. And w.r.t network.get_layers_by_name maybe :

self.all_layers = layer.all_layers
...
self.all_layers[name] = self.output

and:

def get_layers_by_name(self, name):
    print("  Get layers in a network with %s" % name)
    d_layers = self.all_layers.get(name, None)
    if d_layers is None:
        print(" %s not implemented ".format(name))
    return d_layers