Master's Project Evaluation

This repository contains code to reproduce the results of my Master's Project.

Getting Started

Use conda to create the required Python environment and activate it:

$ conda env create -f environment.yml
$ conda activate masters-proj-eval

Scripts

• scripts/eval_image_reconstruction.py: Evaluate non-blind deblurring or single image super-resolution on one dataset for one model. See --help for parameters.
• scripts/eval_image_reconstruction_all.py: Evaluate nb deblurring and sisr on a set of datasets and a folder of models. See --help for parameters.
• scripts/find_noise_stddev_for_dmsp.py: Evaluate DMSP on a range of different noise stddev values used for the prior. See --help for parameters.

NOTE: For now, use the evaluate.py script in HedghogCode/denosing-gan for denoising evaluation. An adapted version of this script will be added to this repository soon.

Notebooks

The folder notebooks/ contains Jupyter notebooks.

• visualize_*.ipynb: Runs reconstruction on one image and saves the results such that they can be included in the report.
• evaluate_*.ipynb: Reads the results from the appropriate scripts and formats them in a table.

Other Methods

The folder other_methods/ contains code to reproduce the results reported in the report for other methods. See the README.md file in the appropriate folder for instructions.

Model Converters

The folder model_converters/ contains scripts for converting existing pretrained models to TensorFlow h5 models.