UNet based Deblurring

Setting

The Kaggle Dogs vs. Cats dataset, originally designed for binary classification, was repurposed to create a robust image restoration framework.

By artificially introducing Gaussian blur to these high-quality pet images, I established a controlled environment for training and evaluating deblurring algorithms.

This strategic adaptation of the dataset is particularly compelling as it provides a diverse collection of natural images with rich textures, complex features, and varying lighting conditions

---

Setup

This code has been tested with Python 3.8.8, Torch 1.10.0

• Setup requirements

pip install -r requirements.txt

---

Public Datasets

Dogs vs. Cats : https://www.kaggle.com/c/dogs-vs-cats

---

Start Training

python train.py