Input layer didn't get converted

[AIGirl10]

Trying to convert keras model to NCNN.

When tried to convert model, it got converted successfully but convolution layer was the first layer

Convolution              conv2d_3                                0 1  conv2d_3_blob 0=16 1=3 2=1 3=1 4=0 5=1 6=432 9=1 11=3 12=1 13=1 

Not able to understand why input layer (data) couldn't get converted. Any reasons ?

Please help ! thank you.

@MarsTechHAN

[MarsTechHAN]

Hi, the input layer should be converted. The converter should find the InputLayer layer and convert that to ncnn Input layer.
Can you attach your keras file and a screenshot of the file opened by Netron.

[AIGirl10]

@MarsTechHAN

Model link: here

Netron Graph: here

[MarsTechHAN]

Hi, from the decoded model config string, there is no InputLayer inside your model. Without an input layer, there is no way to know your input shape settings.

However, I am thinking maybe I should add a default input Layer. I will PR that later.

If you want to fix your model. you can edit your model config to:

7767517
12 12
Input                    input_0                                 0 1 input_0_blob
Convolution              conv2d_3                                1 1 input_0_blob conv2d_3_blob 0=16 1=3 2=1 3=1 4=0 5=1 6=432 9=1 11=3 12=1 13=1 
Pooling                  average_pooling2d_3                     1 1 conv2d_3_blob average_pooling2d_3_blob 0=1 1=2 11=2 2=2 12=2 3=0 4=0 5=1 
Convolution              conv2d_4                                1 1 average_pooling2d_3_blob conv2d_4_blob 0=32 1=3 2=1 3=1 4=0 5=1 6=4608 9=1 11=3 12=1 13=1 
Pooling                  average_pooling2d_4                     1 1 conv2d_4_blob average_pooling2d_4_blob 0=1 1=2 11=2 2=2 12=2 3=0 4=0 5=1 
Convolution              conv2d_5                                1 1 average_pooling2d_4_blob conv2d_5_blob 0=64 1=3 2=1 3=1 4=0 5=1 6=18432 9=1 11=3 12=1 13=1 
Pooling                  average_pooling2d_5                     1 1 conv2d_5_blob average_pooling2d_5_blob 0=1 1=2 11=2 2=2 12=2 3=0 4=0 5=1 
Reshape                  flatten_1                               1 1 average_pooling2d_5_blob flatten_1_blob 0=-1 1=-233 2=-233 3=1 
InnerProduct             dense_3                                 1 1 flatten_1_blob dense_3_blob 0=256 1=1 2=589824 9=1 
InnerProduct             dense_4                                 1 1 dense_3_blob dense_4_blob 0=128 1=1 2=32768 9=1 
InnerProduct             dense_5                                 1 1 dense_4_blob dense_5_blob 0=2 1=1 2=256 9=1 
Softmax                  dense_5_Softmax                         1 1 dense_5_blob dense_5_Softmax_blob 0=0 

---

Here is the decoded config string in your model

{
    "class_name": "Sequential",
    "config": {
        "layers": [
            {
                "class_name": "Conv2D",
                "config": {
                    "activation": "relu",
                    "activity_regularizer": null,
                    "batch_input_shape": [
                        null,
                        64,
                        64,
                        3
                    ],
                    "bias_constraint": null,
                    "bias_initializer": {
                        "class_name": "Zeros",
                        "config": {
                            "dtype": "float32"
                        }
                    },
                    "bias_regularizer": null,
                    "data_format": "channels_last",
                    "dilation_rate": [
                        1,
                        1
                    ],
                    "dtype": "float32",
                    "filters": 16,
                    "kernel_constraint": null,
                    "kernel_initializer": {
                        "class_name": "GlorotUniform",
                        "config": {
                            "dtype": "float32",
                            "seed": null
                        }
                    },
                    "kernel_regularizer": null,
                    "kernel_size": [
                        3,
                        3
                    ],
                    "name": "conv2d_3",
                    "padding": "valid",
                    "strides": [
                        1,
                        1
                    ],
                    "trainable": true,
                    "use_bias": true
                }
            },
            {
                "class_name": "AveragePooling2D",
                "config": {
                    "data_format": "channels_last",
                    "dtype": "float32",
                    "name": "average_pooling2d_3",
                    "padding": "valid",
                    "pool_size": [
                        2,
                        2
                    ],
                    "strides": [
                        2,
                        2
                    ],
                    "trainable": true
                }
            },
            {
                "class_name": "Dropout",
                "config": {
                    "dtype": "float32",
                    "name": "dropout_5",
                    "noise_shape": null,
                    "rate": 0.5,
                    "seed": null,
                    "trainable": true
                }
            },
            {
                "class_name": "Conv2D",
                "config": {
                    "activation": "relu",
                    "activity_regularizer": null,
                    "bias_constraint": null,
                    "bias_initializer": {
                        "class_name": "Zeros",
                        "config": {
                            "dtype": "float32"
                        }
                    },
                    "bias_regularizer": null,
                    "data_format": "channels_last",
                    "dilation_rate": [
                        1,
                        1
                    ],
                    "dtype": "float32",
                    "filters": 32,
                    "kernel_constraint": null,
                    "kernel_initializer": {
                        "class_name": "GlorotUniform",
                        "config": {
                            "dtype": "float32",
                            "seed": null
                        }
                    },
                    "kernel_regularizer": null,
                    "kernel_size": [
                        3,
                        3
                    ],
                    "name": "conv2d_4",
                    "padding": "valid",
                    "strides": [
                        1,
                        1
                    ],
                    "trainable": true,
                    "use_bias": true
                }
            },
            {
                "class_name": "AveragePooling2D",
                "config": {
                    "data_format": "channels_last",
                    "dtype": "float32",
                    "name": "average_pooling2d_4",
                    "padding": "valid",
                    "pool_size": [
                        2,
                        2
                    ],
                    "strides": [
                        2,
                        2
                    ],
                    "trainable": true
                }
            },
            {
                "class_name": "Dropout",
                "config": {
                    "dtype": "float32",
                    "name": "dropout_6",
                    "noise_shape": null,
                    "rate": 0.5,
                    "seed": null,
                    "trainable": true
                }
            },
            {
                "class_name": "Conv2D",
                "config": {
                    "activation": "relu",
                    "activity_regularizer": null,
                    "bias_constraint": null,
                    "bias_initializer": {
                        "class_name": "Zeros",
                        "config": {
                            "dtype": "float32"
                        }
                    },
                    "bias_regularizer": null,
                    "data_format": "channels_last",
                    "dilation_rate": [
                        1,
                        1
                    ],
                    "dtype": "float32",
                    "filters": 64,
                    "kernel_constraint": null,
                    "kernel_initializer": {
                        "class_name": "GlorotUniform",
                        "config": {
                            "dtype": "float32",
                            "seed": null
                        }
                    },
                    "kernel_regularizer": null,
                    "kernel_size": [
                        3,
                        3
                    ],
                    "name": "conv2d_5",
                    "padding": "valid",
                    "strides": [
                        1,
                        1
                    ],
                    "trainable": true,
                    "use_bias": true
                }
            },
            {
                "class_name": "AveragePooling2D",
                "config": {
                    "data_format": "channels_last",
                    "dtype": "float32",
                    "name": "average_pooling2d_5",
                    "padding": "valid",
                    "pool_size": [
                        2,
                        2
                    ],
                    "strides": [
                        2,
                        2
                    ],
                    "trainable": true
                }
            },
            {
                "class_name": "Dropout",
                "config": {
                    "dtype": "float32",
                    "name": "dropout_7",
                    "noise_shape": null,
                    "rate": 0.5,
                    "seed": null,
                    "trainable": true
                }
            },
            {
                "class_name": "Flatten",
                "config": {
                    "data_format": "channels_last",
                    "dtype": "float32",
                    "name": "flatten_1",
                    "trainable": true
                }
            },
            {
                "class_name": "Dense",
                "config": {
                    "activation": "relu",
                    "activity_regularizer": null,
                    "bias_constraint": null,
                    "bias_initializer": {
                        "class_name": "Zeros",
                        "config": {
                            "dtype": "float32"
                        }
                    },
                    "bias_regularizer": null,
                    "dtype": "float32",
                    "kernel_constraint": null,
                    "kernel_initializer": {
                        "class_name": "GlorotUniform",
                        "config": {
                            "dtype": "float32",
                            "seed": null
                        }
                    },
                    "kernel_regularizer": null,
                    "name": "dense_3",
                    "trainable": true,
                    "units": 256,
                    "use_bias": true
                }
            },
            {
                "class_name": "Dropout",
                "config": {
                    "dtype": "float32",
                    "name": "dropout_8",
                    "noise_shape": null,
                    "rate": 0.2,
                    "seed": null,
                    "trainable": true
                }
            },
            {
                "class_name": "Dense",
                "config": {
                    "activation": "relu",
                    "activity_regularizer": null,
                    "bias_constraint": null,
                    "bias_initializer": {
                        "class_name": "Zeros",
                        "config": {
                            "dtype": "float32"
                        }
                    },
                    "bias_regularizer": null,
                    "dtype": "float32",
                    "kernel_constraint": null,
                    "kernel_initializer": {
                        "class_name": "GlorotUniform",
                        "config": {
                            "dtype": "float32",
                            "seed": null
                        }
                    },
                    "kernel_regularizer": null,
                    "name": "dense_4",
                    "trainable": true,
                    "units": 128,
                    "use_bias": true
                }
            },
            {
                "class_name": "Dense",
                "config": {
                    "activation": "softmax",
                    "activity_regularizer": null,
                    "bias_constraint": null,
                    "bias_initializer": {
                        "class_name": "Zeros",
                        "config": {
                            "dtype": "float32"
                        }
                    },
                    "bias_regularizer": null,
                    "dtype": "float32",
                    "kernel_constraint": null,
                    "kernel_initializer": {
                        "class_name": "GlorotUniform",
                        "config": {
                            "dtype": "float32",
                            "seed": null
                        }
                    },
                    "kernel_regularizer": null,
                    "name": "dense_5",
                    "trainable": true,
                    "units": 2,
                    "use_bias": true
                }
            }
        ],
        "name": "sequential_1"
    }
}

[AIGirl10]

@MarsTechHAN

Ohh okk Thanks. Since netron showed input layer I thought it's there.

About input size model needs 64×64 . Also adding input in Param file doesn't it affect the number of parameters (first line in .param file)

So maybe correct .Param can be something like this

Input                    input_0                                 0 1 input_0_blob 0=1 1=3 2=64 3=64

Right ?

Again Thanks for your help.

[MarsTechHAN]

You can check the code, the first line is just a constant magic, but the second line is the layer count and blob count which should be increment. ncnn uses dynamic shapes alone for all executions, so you don't need to give a constant input shape here.

keras2ncnn/keras2ncnn/ncnn_emitter.py

Lines 35 to 84 in 39f2cb9

	def emit_param(self, file_name, seq):
	ncnn_param_file = open(file_name, 'w+')
	ncnn_param_file.write('%d\n' % self.MAGGGGGIC)
	param_contect = ''
	blob_count = 0
	for layer_name in seq:
	layer_type = self.ncnn_graph.get_node_attr(layer_name)['type']
	input_count = len(self.ncnn_graph.get_node_inbounds(layer_name))
	output_count = len(self.ncnn_graph.get_node_outbounds(layer_name))
	output_count = 1 if output_count == 0 else output_count
	input_blobs = []
	inbound_nodes = self.ncnn_graph.get_node_inbounds(layer_name)
	for in_node in inbound_nodes:
	if len(self.ncnn_graph.get_node_outbounds(in_node)) > 1:
	input_blobs.append(
	'%s_blob_idx_%d' %
	(in_node, self.ncnn_graph.get_node_outbounds(in_node).index(layer_name)))
	else:
	input_blobs.append('%s_blob' % in_node)
	output_blobs = []
	if output_count > 1:
	for i in range(output_count):
	output_blobs.append('%s_blob_idx_%d' % (layer_name, i))
	else:
	output_blobs.append('%s_blob' % layer_name)
	blob_count += len(output_blobs)
	param_contect += (
	('%s' + (
	25 - len(layer_type)) * ' ' + '%s' + (
	40 - len(layer_name)) * ' ' + '%d %d %s %s %s\n') % (layer_type,
	layer_name,
	input_count,
	output_count,
	' '.join(input_blobs),
	' '.join(output_blobs),
	self.ncnn_graph.get_node_attr(layer_name)['param']))
	layer_count = len(self.ncnn_graph.get_graph())
	ncnn_param_file.write('%d %d\n' % (layer_count, blob_count))
	ncnn_param_file.write(param_contect)
	ncnn_param_file.close()

[MarsTechHAN]

Fix in 9841e26, I will close this issue since the issue has been resolved and no other issue raised.