Mastering Computer Vision with TensorFlow 2.x

Iteration 2 – CNN parameter adjustment

Drop one Conv2D 64 and one Conv2D 128 so that the CNN has only one Conv2D 64 and one Conv2D 128 with this change. Also, convert average pooling into max pooling, as shown here:

Epoch 1/5 256/256 [========================] - 63s 247ms/step - loss: 1.7471 - acc: 0.2804 Epoch 2/5 256/256 [========================] - 64s 252ms/step - loss: 1.4631 - acc: 0.4307 Epoch 3/5 256/256 [========================] - 66s 256ms/step - loss: 1.3042 - acc: 0.4990 Epoch 4/5 256/256 [========================] - 66s 257ms/step - loss: 1.2183 - acc: 0.5360 Epoch 5/5 256/256 [========================] - 67s 262ms/step - loss: 1.1407 - acc: 0.5691

Result: Again, this change has no significant effect on accuracy.

Next, we will ...