預(yù)測實現(xiàn)

在上一篇文章中實現(xiàn)了裝甲板id識別的網(wǎng)絡(luò)訓(xùn)練并保存為了ckpt文件
http://www.reibang.com/p/191337a9a819
雖然全連接的網(wǎng)絡(luò)精度也就那樣了礁蔗，但是還是練習(xí)一下用現(xiàn)有的網(wǎng)絡(luò)進行裝甲板id預(yù)測

• 復(fù)現(xiàn)網(wǎng)絡(luò)

#網(wǎng)絡(luò)搭建
model = tf.keras.models.Sequential([
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(500,activation='relu',kernel_regularizer=tf.keras.regularizers.l2()),
    tf.keras.layers.Dense(128,activation='relu',kernel_regularizer=tf.keras.regularizers.l2()),
    tf.keras.layers.Dense(50,activation='relu',kernel_regularizer=tf.keras.regularizers.l2()),
    tf.keras.layers.Dense(8,activation='softmax',kernel_regularizer=tf.keras.regularizers.l2())
])

• 加載參數(shù)

#加載參數(shù)
ckpt_path = "./checkpoint/armor_id.ckpt"
if(os.path.exists(ckpt_path + ".index")):
    print("--load modle--")
    model.load_weights(ckpt_path)
else:
    print('----------------------------------------------error')

• 輸入數(shù)據(jù)處理

#圖片讀取與處理
img = tf.io.read_file (test_img_path)
img_raw = tf.image.decode_bmp (img)
img_raw = tf.cast(img_raw,dtype=tf.float32)
x_predict = tf.convert_to_tensor(img_raw)
x_predict = tf.reshape(x_predict,[1,-1])

• 代碼整體實現(xiàn)

import tensorflow as tf
import os

if __name__ == '__main__':
    test_img_path = './armor_dataset/8/8_47.bmp'
    
    #網(wǎng)絡(luò)搭建
    model = tf.keras.models.Sequential([
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(500,activation='relu',kernel_regularizer=tf.keras.regularizers.l2()),
        tf.keras.layers.Dense(128,activation='relu',kernel_regularizer=tf.keras.regularizers.l2()),
        tf.keras.layers.Dense(50,activation='relu',kernel_regularizer=tf.keras.regularizers.l2()),
        tf.keras.layers.Dense(8,activation='softmax',kernel_regularizer=tf.keras.regularizers.l2())
    ])
    #加載參數(shù)
    ckpt_path = "./checkpoint/armor_id.ckpt"
    if(os.path.exists(ckpt_path + ".index")):
        print("--load modle--")
        model.load_weights(ckpt_path)
    else:
        print('----------------------------------------------error')
        
    #圖片讀取與處理
    img = tf.io.read_file (test_img_path)
    img_raw = tf.image.decode_bmp (img)
    img_raw = tf.cast(img_raw,dtype=tf.float32)
    x_predict = tf.convert_to_tensor(img_raw)
    x_predict = tf.reshape(x_predict,[1,-1])
    
    #預(yù)測結(jié)果
    result = model.predict(x_predict)
    pred = tf.argmax(result,axis=1)   #獲取概率最大數(shù)值的下標
    pred = pred + 1
    print("預(yù)測id為:")
    tf.print(pred)

遇到的坑

• 實際未讀入數(shù)據(jù)
  現(xiàn)象是每次輸出結(jié)果隨機變化
• 使用tfrecord解碼的數(shù)據(jù)和使用原始數(shù)據(jù)解碼的數(shù)據(jù)不一致
  應(yīng)當檢查編碼解碼過程中的類型轉(zhuǎn)換
  http://www.reibang.com/p/51659ec687f8

測試結(jié)果

測試圖片

51.png

5177.png

847.png

還進行了其他數(shù)字的測試
測試圖片基本都實現(xiàn)了正確的預(yù)測