Mastering Computer Vision with TensorFlow 2.x

Iteration 4 – CNN parameter adjustment

In this iteration, all the parameters are the same as they were in the previous iteration, except for strides = 2, which is added after the first Conv2D.

Next, we keep everything the same but add a pooling layer after the first Conv2D:

model.add(Conv2D(64, (3, 3), activation='relu', input_shape=(48,48,1)))model.add(AveragePooling2D(pool_size=(3,3), strides=(2, 2)))model.add(Conv2D(128, (3, 3), activation='relu'))model.add(AveragePooling2D(pool_size=(3,3), strides=(2, 2)))

In this iteration, no change is made to the dense layer. The calculation time is faster but the accuracy has dropped:

Epoch 1/5 256/256 [========================] - 99s 386ms/step - loss: 1.6855 - acc: 0.3240 Epoch 2/5 256/256 [========================] ...