Inception Modules with Connected Local and Global Features implemented with CAM.
ILGNet is a novel neural network to solve the challenging problem of aesthetic image classification. The orininal architecture is given -
We added used class activation methods to predict which region is responsible for aesthetic image classification. To suit the needs of CAMs we modified the network to make it better. We trained this network on AVA dataset and the weights are stores in a .pth file.
An example of the output-
Original image -
Output Heat map + image overlay -
