This model is for recognizing 20 different classes of leaves from 28x28 RGB images.
The model architecture seems somewhat strange compared to typical classification architectures. What is the purpose of the sigmoid layer applied directly on the input?
1 Like
Additionally to @eqy’s comment it also is uncommon to avoid using non-linear activations between the trainable layers.
PS: you can post code snippets by wrapping them into three backticks ```, which makes debugging easier and allows the search engine to index the code.