前言:
經歷了三四天的訓練，總算讓模型維持在9成以上的準確率?
val_accuracy 的部分，我還在想如何提高，有可能需要捨棄掉每次訓練 accuracy 下降的結果
改成只儲存數值高的 accuracy
訓練:

835/835 [==============================] - 367s 439ms/step - batch: 417.0000 - size: 8.0000 - loss: 0.0250 - accuracy: 0.9907 - val_loss: 4.0026 
- val_accuracy: 0.6502
儲存訓練模型
835/835 [==============================] - 368s 440ms/step - batch: 417.0000 - size: 8.0000 - loss: 0.0235 - accuracy: 0.9924 - val_loss: 3.9955 
- val_accuracy: 0.6466
儲存訓練模型
835/835 [==============================] - 369s 442ms/step - batch: 417.0000 - size: 8.0000 - loss: 0.0224 - accuracy: 0.9925 - val_loss: 4.1223 
- val_accuracy: 0.6418
儲存訓練模型
835/835 [==============================] - 336s 402ms/step - batch: 417.0000 - size: 8.0000 - loss: 0.0226 - accuracy: 0.9916 - val_loss: 3.8956 
- val_accuracy: 0.6611
儲存訓練模型
835/835 [==============================] - 326s 390ms/step - batch: 417.0000 - size: 8.0000 - loss: 0.0232 - accuracy: 0.9928 - val_loss: 3.9565 
- val_accuracy: 0.6514
儲存訓練模型

辨識:

python predict_resnet50.py 50.jpg
['50.jpg']
=============================
...
50.jpg
準確率: 99.84%   050.Chinese_shar-pei
準確率: 0.16%   088.Irish_water_spaniel
準確率: 0.00%   003.Airedale_terrier
準確率: 0.00%   069.French_bulldog
準確率: 0.00%   047.Chesapeake_bay_retriever
準確率: 0.00%   051.Chow_chow
準確率: 0.00%   103.Mastiff
準確率: 0.00%   038.Brussels_griffon
準確率: 0.00%   027.Bloodhound
準確率: 0.00%   041.Bullmastiff