ameya105 / trash-cnn

A convolutional neural network on keras for garbage image classification

trash-cnn

This is a convolutional neural network implementation for garbage image classification

The dataset we are using is from the work by Gary Thung and Mindy Yang and can be found here
I have splitted the dataset into train and test set with the following directory structure:

.
+-- dataset-splitted
|   +-- test-set
|   |   +-- cardboard
|   |   +-- glass
|   |   +-- ...
|   +-- training-set
|   |   +-- cardboard
|   |   +-- glass
|   |   +-- ...

Dependencies

• tensorflow (or tensorflow-gpu if using GPU for training)
• keras

Execute

I suggest to run using the -i interpreter option to run commands on terminal after execution such as fit(epochs) or print_layers()

The model

For this project I tried to take advantage of the transfer learning technique which turns out to be very promising. For the notes about the learning with different parameters see: densenet and VGG-19