2. 程式人生 > 程式設計 >keras實現呼叫自己訓練的模型,並去掉全連線層

keras實現呼叫自己訓練的模型,並去掉全連線層

阿新 • • 發佈：2020-06-10

其實很簡單

from keras.models import load_model

base_model = load_model('model_resenet.h5')#載入指定的模型
print(base_model.summary())#輸出網路的結構圖

這是我的網路模型的輸出，其實就是它的結構圖 

__________________________________________________________________________________________________
Layer (type)          Output Shape     Param #   Connected to           
==================================================================================================
input_1 (InputLayer)      (None,227,1) 0                      
__________________________________________________________________________________________________
conv2d_1 (Conv2D)        (None,225,32) 320     input_1[0][0]          
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None,32) 128     conv2d_1[0][0]          
__________________________________________________________________________________________________
activation_1 (Activation)    (None,32) 0      batch_normalization_1[0][0]   
__________________________________________________________________________________________________
conv2d_2 (Conv2D)        (None,32) 9248    activation_1[0][0]        
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None,32) 128     conv2d_2[0][0]          
__________________________________________________________________________________________________
activation_2 (Activation)    (None,32) 0      batch_normalization_2[0][0]   
__________________________________________________________________________________________________
conv2d_3 (Conv2D)        (None,32) 9248    activation_2[0][0]        
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None,32) 128     conv2d_3[0][0]          
__________________________________________________________________________________________________
merge_1 (Merge)         (None,32) 0      batch_normalization_3[0][0]   
                                 activation_1[0][0]        
__________________________________________________________________________________________________
activation_3 (Activation)    (None,32) 0      merge_1[0][0]          
__________________________________________________________________________________________________
conv2d_4 (Conv2D)        (None,32) 9248    activation_3[0][0]        
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None,32) 128     conv2d_4[0][0]          
__________________________________________________________________________________________________
activation_4 (Activation)    (None,32) 0      batch_normalization_4[0][0]   
__________________________________________________________________________________________________
conv2d_5 (Conv2D)        (None,32) 9248    activation_4[0][0]        
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None,32) 128     conv2d_5[0][0]          
__________________________________________________________________________________________________
merge_2 (Merge)         (None,32) 0      batch_normalization_5[0][0]   
                                 activation_3[0][0]        
__________________________________________________________________________________________________
activation_5 (Activation)    (None,32) 0      merge_2[0][0]          
__________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None,112,32) 0      activation_5[0][0]        
__________________________________________________________________________________________________
conv2d_6 (Conv2D)        (None,110,64) 18496    max_pooling2d_1[0][0]      
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None,64) 256     conv2d_6[0][0]          
__________________________________________________________________________________________________
activation_6 (Activation)    (None,64) 0      batch_normalization_6[0][0]   
__________________________________________________________________________________________________
conv2d_7 (Conv2D)        (None,64) 36928    activation_6[0][0]        
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None,64) 256     conv2d_7[0][0]          
__________________________________________________________________________________________________
activation_7 (Activation)    (None,64) 0      batch_normalization_7[0][0]   
__________________________________________________________________________________________________
conv2d_8 (Conv2D)        (None,64) 36928    activation_7[0][0]        
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None,64) 256     conv2d_8[0][0]          
__________________________________________________________________________________________________
merge_3 (Merge)         (None,64) 0      batch_normalization_8[0][0]   
                                 activation_6[0][0]        
__________________________________________________________________________________________________
activation_8 (Activation)    (None,64) 0      merge_3[0][0]          
__________________________________________________________________________________________________
conv2d_9 (Conv2D)        (None,64) 36928    activation_8[0][0]        
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None,64) 256     conv2d_9[0][0]          
__________________________________________________________________________________________________
activation_9 (Activation)    (None,64) 0      batch_normalization_9[0][0]   
__________________________________________________________________________________________________
conv2d_10 (Conv2D)       (None,64) 36928    activation_9[0][0]        
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None,64) 256     conv2d_10[0][0]         
__________________________________________________________________________________________________
merge_4 (Merge)         (None,64) 0      batch_normalization_10[0][0]   
                                 activation_8[0][0]        
__________________________________________________________________________________________________
activation_10 (Activation)   (None,64) 0      merge_4[0][0]          
__________________________________________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None,55,64)  0      activation_10[0][0]       
__________________________________________________________________________________________________
conv2d_11 (Conv2D)       (None,53,64)  36928    max_pooling2d_2[0][0]      
__________________________________________________________________________________________________
batch_normalization_11 (BatchNo (None,64)  256     conv2d_11[0][0]         
__________________________________________________________________________________________________
activation_11 (Activation)   (None,64)  0      batch_normalization_11[0][0]   
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None,26,64)  0      activation_11[0][0]       
__________________________________________________________________________________________________
conv2d_12 (Conv2D)       (None,64)  36928    max_pooling2d_3[0][0]      
__________________________________________________________________________________________________
batch_normalization_12 (BatchNo (None,64)  256     conv2d_12[0][0]         
__________________________________________________________________________________________________
activation_12 (Activation)   (None,64)  0      batch_normalization_12[0][0]   
__________________________________________________________________________________________________
conv2d_13 (Conv2D)       (None,64)  36928    activation_12[0][0]       
__________________________________________________________________________________________________
batch_normalization_13 (BatchNo (None,64)  256     conv2d_13[0][0]         
__________________________________________________________________________________________________
merge_5 (Merge)         (None,64)  0      batch_normalization_13[0][0]   
                                 max_pooling2d_3[0][0]      
__________________________________________________________________________________________________
activation_13 (Activation)   (None,64)  0      merge_5[0][0]          
__________________________________________________________________________________________________
conv2d_14 (Conv2D)       (None,64)  36928    activation_13[0][0]       
__________________________________________________________________________________________________
batch_normalization_14 (BatchNo (None,64)  256     conv2d_14[0][0]         
__________________________________________________________________________________________________
activation_14 (Activation)   (None,64)  0      batch_normalization_14[0][0]   
__________________________________________________________________________________________________
conv2d_15 (Conv2D)       (None,64)  36928    activation_14[0][0]       
__________________________________________________________________________________________________
batch_normalization_15 (BatchNo (None,64)  256     conv2d_15[0][0]         
__________________________________________________________________________________________________
merge_6 (Merge)         (None,64)  0      batch_normalization_15[0][0]   
                                 activation_13[0][0]       
__________________________________________________________________________________________________
activation_15 (Activation)   (None,64)  0      merge_6[0][0]          
__________________________________________________________________________________________________
max_pooling2d_4 (MaxPooling2D) (None,13,64)  0      activation_15[0][0]       
__________________________________________________________________________________________________
conv2d_16 (Conv2D)       (None,11,32)  18464    max_pooling2d_4[0][0]      
__________________________________________________________________________________________________
batch_normalization_16 (BatchNo (None,32)  128     conv2d_16[0][0]         
__________________________________________________________________________________________________
activation_16 (Activation)   (None,32)  0      batch_normalization_16[0][0]   
__________________________________________________________________________________________________
conv2d_17 (Conv2D)       (None,32)  9248    activation_16[0][0]       
__________________________________________________________________________________________________
batch_normalization_17 (BatchNo (None,32)  128     conv2d_17[0][0]         
__________________________________________________________________________________________________
activation_17 (Activation)   (None,32)  0      batch_normalization_17[0][0]   
__________________________________________________________________________________________________
conv2d_18 (Conv2D)       (None,32)  9248    activation_17[0][0]       
__________________________________________________________________________________________________
batch_normalization_18 (BatchNo (None,32)  128     conv2d_18[0][0]         
__________________________________________________________________________________________________
merge_7 (Merge)         (None,32)  0      batch_normalization_18[0][0]   
                                 activation_16[0][0]       
__________________________________________________________________________________________________
activation_18 (Activation)   (None,32)  0      merge_7[0][0]          
__________________________________________________________________________________________________
conv2d_19 (Conv2D)       (None,32)  9248    activation_18[0][0]       
__________________________________________________________________________________________________
batch_normalization_19 (BatchNo (None,32)  128     conv2d_19[0][0]         
__________________________________________________________________________________________________
activation_19 (Activation)   (None,32)  0      batch_normalization_19[0][0]   
__________________________________________________________________________________________________
conv2d_20 (Conv2D)       (None,32)  9248    activation_19[0][0]       
__________________________________________________________________________________________________
batch_normalization_20 (BatchNo (None,32)  128     conv2d_20[0][0]         
__________________________________________________________________________________________________
merge_8 (Merge)         (None,32)  0      batch_normalization_20[0][0]   
                                 activation_18[0][0]       
__________________________________________________________________________________________________
activation_20 (Activation)   (None,32)  0      merge_8[0][0]          
__________________________________________________________________________________________________
max_pooling2d_5 (MaxPooling2D) (None,5,32)   0      activation_20[0][0]       
__________________________________________________________________________________________________
conv2d_21 (Conv2D)       (None,3,64)   18496    max_pooling2d_5[0][0]      
__________________________________________________________________________________________________
batch_normalization_21 (BatchNo (None,64)   256     conv2d_21[0][0]         
__________________________________________________________________________________________________
activation_21 (Activation)   (None,64)   0      batch_normalization_21[0][0]   
__________________________________________________________________________________________________
conv2d_22 (Conv2D)       (None,64)   36928    activation_21[0][0]       
__________________________________________________________________________________________________
batch_normalization_22 (BatchNo (None,64)   256     conv2d_22[0][0]         
__________________________________________________________________________________________________
activation_22 (Activation)   (None,64)   0      batch_normalization_22[0][0]   
__________________________________________________________________________________________________
conv2d_23 (Conv2D)       (None,64)   36928    activation_22[0][0]       
__________________________________________________________________________________________________
batch_normalization_23 (BatchNo (None,64)   256     conv2d_23[0][0]         
__________________________________________________________________________________________________
merge_9 (Merge)         (None,64)   0      batch_normalization_23[0][0]   
                                 activation_21[0][0]       
__________________________________________________________________________________________________
activation_23 (Activation)   (None,64)   0      merge_9[0][0]          
__________________________________________________________________________________________________
conv2d_24 (Conv2D)       (None,64)   36928    activation_23[0][0]       
__________________________________________________________________________________________________
batch_normalization_24 (BatchNo (None,64)   256     conv2d_24[0][0]         
__________________________________________________________________________________________________
activation_24 (Activation)   (None,64)   0      batch_normalization_24[0][0]   
__________________________________________________________________________________________________
conv2d_25 (Conv2D)       (None,64)   36928    activation_24[0][0]       
__________________________________________________________________________________________________
batch_normalization_25 (BatchNo (None,64)   256     conv2d_25[0][0]         
__________________________________________________________________________________________________
merge_10 (Merge)        (None,64)   0      batch_normalization_25[0][0]   
                                 activation_23[0][0]       
__________________________________________________________________________________________________
activation_25 (Activation)   (None,64)   0      merge_10[0][0]          
__________________________________________________________________________________________________
max_pooling2d_6 (MaxPooling2D) (None,1,64)   0      activation_25[0][0]       
__________________________________________________________________________________________________
flatten_1 (Flatten)       (None,64)      0      max_pooling2d_6[0][0]      
__________________________________________________________________________________________________
dense_1 (Dense)         (None,256)     16640    flatten_1[0][0]         
__________________________________________________________________________________________________
dropout_1 (Dropout)       (None,256)     0      dense_1[0][0]          
__________________________________________________________________________________________________
dense_2 (Dense)         (None,2)      514     dropout_1[0][0]         
==================================================================================================
Total params: 632,098
Trainable params: 629,538
Non-trainable params: 2,560
__________________________________________________________________________________________________

去掉模型的全連線層

from keras.models import load_model

base_model = load_model('model_resenet.h5')
resnet_model = Model(inputs=base_model.input,outputs=base_model.get_layer('max_pooling2d_6').output)
#'max_pooling2d_6'其實就是上述網路中全連線層的前面一層，當然這裡你也可以選取其它層，把該層的名稱代替'max_pooling2d_6'即可，這樣其實就是擷取網路，輸出網路結構就是方便讀取每層的名字。
print(resnet_model.summary())

新輸出的網路結構： 

__________________________________________________________________________________________________
Layer (type)          Output Shape     Param #   Connected to           
==================================================================================================
input_1 (InputLayer)      (None,64)   0      activation_25[0][0]       
==================================================================================================
Total params: 614,944
Trainable params: 612,384
Non-trainable params: 2,560
__________________________________________________________________________________________________

以上這篇keras實現呼叫自己訓練的模型,並去掉全連線層就是小編分享給大家的全部內容了，希望能給大家一個參考，也希望大家多多支援我們。

相關推薦

keras實現呼叫自己訓練的模型,並去掉全連線層

其實很簡單 from keras.models import load_model base_model = load_model(\'model_resenet.h5\')#載入指定的模型

Keras 實現載入預訓練模型並凍結網路的層

在解決一個任務時，我會選擇載入預訓練模型並逐步fine-tune。比如，分類任務中，優異的深度學習網路有很多。

pytorch 實現在預訓練模型的 input上增減通道

如何把imagenet預訓練的模型，輸入層的通道數隨心所欲的修改，從而來適應自己的任務

Keras實現將兩個模型連線到一起

神經網路玩得越久就越會嘗試一些網路結構上的大改動。 先說意圖 有兩個模型：模型A和模型B。模型A的輸出可以連線B的輸入。將兩個小模型連線成一個大模型，A-B，既可以同時訓練又可以分離訓練。

使用Keras實現簡單線性迴歸模型操作

神經網路可以用來模擬迴歸問題 (regression)，實質上是單輸入單輸出神經網路模型，例如給下面一組資料，用一條線來對資料進行擬合，並可以預測新輸入 x 的輸出值。

淺談keras通過model.fit_generator訓練模型(節省記憶體)

前言 前段時間在訓練模型的時候，發現當訓練集的數量過大，並且輸入的圖片維度過大時，很容易就超記憶體了，舉個簡單例子，如果我們有20000個樣本，輸入圖片的維度是224x224x3，用float32儲存，那麼如果我們一次性將

從零開始開發自己的類keras深度學習框架2 ：實現全連線層

技術標籤：深度學習神經網路神經網路深度學習 認真學習，佛系更博。 上一章簡單介紹瞭如何實現資料的讀取功能，本章將詳細介紹如何實現神經網路最基礎的層：全連線層。

gensim訓練模型並使用以人民的名義小說為例子

首先對小說進行預處理，使用jieba分詞進行分詞 import jieba import jieba.analyse jieba.suggest_freq(\'沙瑞金\', True)

Pytorch修改ResNet模型全連線層進行直接訓練例項

之前在用預訓練的ResNet的模型進行遷移訓練時，是固定除最後一層的前面層權重，然後把全連線層輸出改為自己需要的數目，進行最後一層的訓練，那麼現在假如想要只是把

tensorflow2.0——自定義全連線層實現並儲存

import tensorflow as tf def preprocess(x, y): x = tf.cast(x, dtype=tf.float32) / 255 - 0.5 y = tf.cast(y, dtype=tf.int32)

pytorch三層全連線層實現手寫字母識別方式

先用最簡單的三層全連線神經網路，然後新增啟用層檢視實驗結果，最後加上批標準化驗證是否有效

TensorFlow實現儲存訓練模型為pd檔案並恢復

TensorFlow儲存模型程式碼 import tensorflow as tf from tensorflow.python.framework import graph_util

Python實現Keras搭建神經網路訓練分類模型教程

我就廢話不多說了，大家還是直接看程式碼吧~ 註釋講解版： # Classifier example import numpy as np

keras實現theano和tensorflow訓練的模型相互轉換

我就廢話不多說了，大家還是直接看程式碼吧~ </pre><pre code_snippet_id=\"1947416\" snippet_file_name=\"blog_20161025_1_3331239\" name=\"code\" class=\"python\">

使用Keras建立模型並訓練等一系列操作方式

由於Keras是一種建立在已有深度學習框架上的二次框架，其使用起來非常方便，其後端實現有兩種方法，theano和tensorflow。由於自己平時用tensorflow，所以選擇後端用tensorflow的Keras，程式碼寫起來更加方便。

使用預先訓練好的模型來訓練小訓練集tf-keras實現

> //20201104 > 寫在前面：最近在練手keras的相關專案，今天做了一個使用預訓練模型訓練小資料集的專案，在此記錄總結一下

golang實現自己的模組並呼叫

官方教程地址：https://golang.google.cn/doc/tutorial/call-module-code 1.在程式碼目錄建立一個目錄greetings 用來存放greetings 模組

pytorch載入預訓練模型後,實現訓練指定層

1、有了已經訓練好的模型引數，對這個模型的某些層做了改變，如何利用這些訓練好的模型引數繼續訓練：

pytorch 實現將自己的圖片資料處理成可以訓練的圖片型別

為了使用自己的影象資料，需要仿照pytorch資料輸入建立新的類，其中資料格式為numpy.ndarray。

python介面呼叫已訓練好的caffe模型測試分類方法

訓練好了model後，可以通過python呼叫caffe的模型，然後進行模型測試的輸出。 本次測試主要依靠的模型是在caffe模型裡面自帶訓練好的結構引數：~/caffe/models/bvlc_reference_caffenet/bvlc_reference_caffenet.caf