Kerras模型的定义在模型的输入张量是另一个模型的输出时发生变化

Question

我需要提供model1的输出作为model2的输入。

A=keras.layers.Input(batch_shape=(3,500,500,3))
B=keras.layers.Conv2D(filters=3, kernel_size=3)(A)
model1=keras.models.Model(A,B)
C=keras.layers.Conv2D(filters=3, kernel_size=3)(B)
model2=keras.applications.vgg19.VGG19(input_tensor=B, weights='imagenet', include_top=False, pooling='avg')
#Error: You are trying to load a weight file containing 16 layers into a model with 17 layers.

正如错误消息所示，keras无法正确实现vgg网络。

vgg19网包含23层（不含顶部）。但如果输入张量是另一个模型的输出，则层数会发生变化。

vgg19_normal=keras.applications.vgg19.VGG19(input_tensor=A, weights=None, include_top=False, pooling='avg')
len(vgg19_normal.layers)
#23

vgg19_abnormal=keras.applications.vgg19.VGG19(input_tensor=B, weights=None, include_top=False, pooling='avg')
#input tensor B is the output of model1
len(vgg19_abnormal.layers)
#24

所以keras抱怨层数不相等。

Answer 1

我仍有一些疑问，例如我不明白 ngOnInit(): void { const that = this; that.dtOptions = { pagingType: 'full_numbers', 'lengthMenu': [[10, 25, 50, -1], [10, 25, 50, 'All']], responsive: true, search: '_INPUT_', searchPlaceholder: 'Search records', pageLength: 2, serverSide: true, processing: true, ajax: (dataTablesParameters: any, callback) => { that.http .post<any[]>( 'api/prodotti/prodotti_executing/', dataTablesParameters, { } ).subscribe(resp => { that.prodotti = resp; callback({ recordsTotal: resp.length, data: [] }); }); }, //columns: [{ data: 'id' }, { data: 'firstName' }, { data: 'lastName' }] }; } ngAfterViewInit() { const table = $('#datatables').DataTable(); // Edit record table.on( 'click', '.edit', function () { const $tr = $(this).closest('tr'); console.log(table); const data = table.row($tr).data(); alert( 'You press on Row: ' + data[0] + ' ' + data[1] + ' ' + data[2] + '\'s row.' ); } ); // Delete a record table.on( 'click', '.remove', function (e: any) { const $tr = $(this).closest('tr'); table.row($tr).remove().draw(); e.preventDefault(); } ); 的用途。无论如何，这是我目前的想法：

C

A=keras.layers.Input(batch_shape=(3,500,500,3)) B=keras.layers.Conv2D(filters=3, kernel_size=3)(A) model1=keras.models.Model(A,B) C=keras.layers.Conv2D(filters=3, kernel_size=3)(B) vgg19=keras.applications.vgg19.VGG19(weights='imagenet', include_top=False, pooling='avg')(model1.output) model2=keras.models.Model(inputs=A, outputs=vgg19) 的结果拓扑是

要制作此图片，我使用了keras.utils.plot_model。