One layer of conv2d and pooling is better than 2 layers? WHY?

Course Q&A TensorFlow Developer Professional Introduction to TF for Artificial Intelligence ...
1

Screen Shot 2022-01-21 at 11.56.21 PM
Screen Shot 2022-01-21 at 11.56.21 PM1750×1098 156 KB

Screen Shot 2022-01-21 at 11.56.31 PM
Screen Shot 2022-01-21 at 11.56.31 PM1776×584 44.8 KB

So the result is 95% and 91% respectively compare to the 1 layer convolution. Any logic behind of it?

2

Hello @Haojun_Song !

Although i don’t have access to the assignement but i can see that you set 64 filters for the first convolution layer and a 16 for the second so what if you try to inverse them and compare the results ?

Hope this will help