This small one-week project was implemented for the Deep Learning course of my master in Artficial Intelligence. It consists in creating a deblurring Deep Neural Network, aimed to remove Gaussian blur and Gaussian noise from images. The dataset is built starting from CIFAR-10 images collection. The noisy version of the images is obtained by applying Gaussian blur and noise to the original images themselves.
The metric used to evaluate the results is the MSE. Moreover, an visual inspection on the results is illustrated in the notebook.
.
├── images
│ ├── example.png # Example of original and blurred images
│ └── model.png # Diagram illustrating the DNN model
├── previous attempts
│ ├── DnCNN.ipynb # Notebook of a previous attempt using a DnCNN model
│ ├── autoencoder_with_subtraction.ipynb # Notebook of a previous attempt using an autoencoder with subtraction
│ └── naive_autoencoder.ipynb # Notebook of a previous attempt using a naive autoencoder
├── weights
│ └── weights.h5 # Weights of the RIDNET DNN
├── RIDNET.ipynb # Main Notebook of the project using a RIDNET model
├── .gitattributes
├── .gitignore
├── LICENSE
└── README.md
Git is used for versioning.
This project is licensed under the MIT License - see the LICENSE file for details.