keras 模型可视化显示卷积conv和池化pooling kernel_size

╰+哭是因爲堅強的太久メ 2022-05-24 03:26 400阅读 0赞

keras 模型可视化可参考:[keras 模型可视化和遇到的坑][keras]

但是默认的方法`plot_model(model, to_file='model_ResNet.png',show_shapes=True)`并不能显示conv和pooling的kernel\_size,所以需要修改一下plot\_model的方法plot\_model->model\_to\_dot。

def plot_model(model, to_file='model.png', show_shapes=False, show_layer_names=True, rankdir='TB'):
        dot = model_to_dot(model, show_shapes, show_layer_names, rankdir)#需要修改的方法，再进入
        _, extension = os.path.splitext(to_file)
        if not extension:
            extension = 'png'
        else:
            extension = extension[1:]
        dot.write(to_file, format=extension)
    
    def model_to_dot(model, show_shapes=False, show_layer_names=True, rankdir='TB'):
        from ..layers.wrappers import Wrapper
        from ..models import Sequential
    
        _check_pydot()
        dot = pydot.Dot()
        dot.set('rankdir', rankdir)
        dot.set('concentrate', True)
        dot.set_node_defaults(shape='record')
    
        if isinstance(model, Sequential):
            if not model.built:
                model.build()
            model = model.model
        layers = model.layers
    
        # Create graph nodes.
        for layer in layers:
            layer_id = str(id(layer))
    
            # Append a wrapped layer's label to node's label, if it exists.
            layer_name = layer.name
            class_name = layer.__class__.__name__
            if isinstance(layer, Wrapper):
                layer_name = '{}({})'.format(layer_name, layer.layer.name)
                child_class_name = layer.layer.__class__.__name__
                class_name = '{}({})'.format(class_name, child_class_name)
    
            # Create node's label.
            #修改的位置，判断一下模型的名称，如果是conv或者pooling，就显示
            if show_layer_names:
                if class_name in ["Conv2D"]:
                    label = '{}: {}'.format(layer_name, class_name+str(layer.kernel_size))
                elif class_name in ["AveragePooling2D","MaxPooling2D"]:
                    label = '{}: {}'.format(layer_name, class_name+str(layer.pool_size))
                else:
                    label = '{}: {}'.format(layer_name, class_name)
            else:
                label = class_name
    
            # Rebuild the label as a table including input/output shapes.
            if show_shapes:
                try:
                    outputlabels = str(layer.output_shape)
                except AttributeError:
                    outputlabels = 'multiple'
                if hasattr(layer, 'input_shape'):
                    inputlabels = str(layer.input_shape)
                elif hasattr(layer, 'input_shapes'):
                    inputlabels = ', '.join(
                        [str(ishape) for ishape in layer.input_shapes])
                else:
                    inputlabels = 'multiple'
                label = '%s\n|{input:|output:}|{ {%s}|{%s}}' % (label,
                                                               inputlabels,
                                                               outputlabels)
            node = pydot.Node(layer_id, label=label)
            dot.add_node(node)
    
        # Connect nodes with edges.
        for layer in layers:
            layer_id = str(id(layer))
            for i, node in enumerate(layer._inbound_nodes):
                node_key = layer.name + '_ib-' + str(i)
                if node_key in model._container_nodes:
                    for inbound_layer in node.inbound_layers:
                        inbound_layer_id = str(id(inbound_layer))
                        layer_id = str(id(layer))
                        dot.add_edge(pydot.Edge(inbound_layer_id, layer_id))
        return dot

修改后运行代码,将ResNet模型可视化:

# coding=utf-8
    ''' @author: admin '''
    from keras.applications.resnet50 import ResNet50
    from keras.utils import plot_model
    
    model = ResNet50(weights='imagenet')
    
    plot_model(model, to_file='model_ResNet1.png',show_shapes=True)

输出ResNet模型图:  
![这里写图片描述][70]

[keras]: https://blog.csdn.net/u011311291/article/details/80298563
[70]: /images/20220524/3f9b67c7c4b34a1381eba43d748ef2e1.png