For the model's architecture, we will use a variation of the VGG network. This architecture is taken from a publicly available script from the forums ( https://www.kaggle.com/jeffd23/dogs-vs-cats-redux-kernels-edition/catdognet-keras-convnet-starter) and here we will adapt this example to DeepLearning4j.
VGG is a model that took 2nd place in the image net 2014 challenge, and it uses only 3 x 3 and 2 x 2 convolutional filters.
It is always a good idea to use the existing architectures, as it solves a lot of time--coming up with a good architecture on your own is a challenging task.
The architecture we will use is as follows:
- Two layers of 3 x 3 convolution with 32 filters
- Max pooling
- Two layers of 3 x 3 convolution with ...
