mdmonoar/self_supervised_deblur

Self Supervised Deblurring

In this project a self supervised deblurring model was built for deblurring blurred images. Then a semantic segmentation model was applied to segment different objects.

Data

A popular dataset named CamVid(Cambridge-Driving Labeled Video Database) was used for this project and was taken from popular data repository Kaggle. It consists of over 700 images and masks for semantic segmentation. The images and masks were seperated in training, validation and testing sets. The ground truth label associate each pixel with one of 32 semantic classes.

Usage

It can be used to dblur blurry images.

Procedure

• A blur function was built which applies random level of gaussian blurring to each image
• All images were blurred using the blur function
• Then the training and validation images were blurred again.This batch was used for training the self-supervised model and the previous batch was as ground truth
• A self supervied deblurring model was built and trained
• Test images were deblurred using the trained model
• The performance of the segmentation model(previously made) was comapred with original test images and newly deblurred images.