1. Read the images into the training set and the testing set ![](https://i.imgur.com/vkrwHWa.png) ![](https://i.imgur.com/c5ZdaLf.png) 2. Design the CNN architecture and train ![](https://i.imgur.com/ow1nOoO.png) 3. 嘗試更改不同的batch_size 最後發現切成1/3 設為180會訓練出比較高的準確度 ![](https://i.imgur.com/gOnBKgM.png) ![](https://i.imgur.com/CBKFc4V.png)