mburakergenc / DNN-Results-on-Negative-Images

DNN-Results-on-Negative-Images (Currently only on CNN)

This repository is created after a discussion on a paper which can be found here.

Thanks to Fuat Beser for his presentation on a group called Derin Ogrenme.

How to use the script

1. Clone the repository and cd into the directory
2. Run pip install --editable . to install the CLI
3. Run the script using trainer --model {architecture-name}
4. Current architectures are

• lenet_5
• mvgg_5
• mvgg_6
• mvgg_7
• mvgg_8
• mvgg_9

5. You can always use trainer --help command to see the options available

##What's done in this repo?

1. Training regular MNIST images with a simple CNN model
2. Converting regular images into negative ones and training the model only on negative images dataset.
3. Combining both regular and negative images and train a model on this combined dataset

##To-Do

1. Improve the CLI.
2. Simply find a way to fix the issue with the negative images! :)