Welcome to AI Blitz XIII! 🚀 | Starter Kit For This Challenge! 🛠

Community Contribution Prizes 📓  |  Find Teammates 👯‍♀️

Discord AI Community 🎧

Overview

How many times have you clicked a cool selfie, only to find out that it was blurred and you could not post it on your Instagram. Successfully solve this puzzle and you could deblur all those blurry selfies!

The goal of this puzzle of single image deblurring is  to recover a clear image from a blurred input image. Deblurring is a difficult task wherein generic methods do not perform well on real-world data. Instead, domain-specific methods for deblurring specific categories like text or face outperform generic counterpart.

🕵️ Problem Statement

The dataset is built on artificially generated human faces that resemble real-world data. Face images are highly structured, composed of several components and semantic information.

In this puzzle, you are given a blurred face image as input data. Your task is to convert the blurred face into a clear image.   

💪 Getting Started

Our Starter Kit helps you understand the submission format by submitting the blurred image as submission in the required format.

💾 Dataset

The dataset is split into 3 different sets - train, validation & test set. The training & validation set will be used in training your models and the predictions generated from the test set will be used to evaluate your model.   Each set contains 5000, 2000, and 3000 samples respectively.

The training and validation set are zip files that contain two folders, blur and original containing blurred and non-blurred images respectively. Both folders will contain the same number of corresponding .jpg image files with dimensions 512x512. For ex.

train.zip						
├── blur
│   ├── f5ka8.jpg	
│   ├── dk5ns.jpg
│   ├── 3knds.jpg
│   └── ...
└── original
    ├── f5ka8.jpg
    ├── dk5ns.jpg
    ├── 3knds.jpg
    └── …

The test set is also a zip file containing only a folder, blur, this folder will contain 3000 jpg images with dimensions 512x512.

test.zip
└── blur
    ├── 3n593.jpg
    ├── sko5d.jpg
    ├── 29sns.jpg
    └── …

📁 Files

Following files are available in the resources section:

• train.zip - ( 5k samples ) This zip file contains the samples for the training set.
• val.zip - ( 2k samples ) This zip file contains the samples for the validation set.
• test.zip - ( 3k samples ) This zip file contains the samples for the testing set without labels.
• sample_submission.zip - This zip file represents how your submission file should look like when making a submission.  

🚀  Submission

Learn to make your first submission using the starter kit 🚀

• Create a submission folder in your working directory,  inside the submission folder create another folder named original.

• Generate all the predictions for test.zip with respective image names with dimensions 512x512 and save them in the folder name original. Make sure you have all 3k images in the folder.  

• Inside a submission directory, put the .ipynb notebook from which you trained the model and generated predictions and save it as notebook.ipynb.

• Zip the submission directory

Overall, this is what your submission directory should look like

submission.zip
├── original
│   ├── 3n593.jpg
│   ├── sko5d.jpg
│   ├── 29sns.jpg
│   └── ...
└── original_notebook.ipynb

Make your first submission here 🚀 !!

🖊 Evaluation Criteria

During the evaluation, the SSIM as the primary score, and PSNR as the secondary score will be used to test the efficiency of the model.

📱 Contact

• Divyanshu Kumar
• Shubhamai