antonioenas / denoising_hpatches_for_descriptor_learning

Imperial College London Deep Learning EE3-25 codes submission repository: descriptor learning on the noisy HPatches dataset.

Hello and welcome to Denoising_HPatches_for_Descriptor_Learning.

This repository contains the software submission for the Deep learning coursework at Imperial College London.

Submission details:

• Name: Antonio Enas
• CID: 01070859
• Login: AE1115

Further instructions:

• All codes in the repository are accompanied by detailed instructions and comments. The content of each code/folder is explained below.

In this folder:

• 1_baseline_denoiser.ipynb This Colaboratory notebook generates, trains and saves the baseline denoiser model.
• 2_baseline_descriptor.ipynb This Colaboratory notebook generates, trains and saves the baseline descriptor model.
• 3_improved_denoiser.ipynb This Colaboratory notebook generates, trains and saves the improved denoiser model (stacked DnCNN).
• 4_improved_descriptor.ipynb This Colaboratory notebook generates, trains and saves the improved descriptor model based on adapted loss function.
• 5_complete_evaluation.ipynb This Colaboratory notebook contains all the codes for qualitative and quantitative analysis and evaluation.
• AE1115.zip This folder contains all the codes above in a downloadable .zip format
• Trained keras models (.h5 files) in the respective .zip folders