This repository contains the DIV2K dataset, which has been compressed using Guetzli for optimized storage. The DIV2K dataset is widely used for image super-resolution and image restoration tasks.
The DIV2K dataset consists of 900 high-quality 2K resolution images that serve as a comprehensive benchmark for evaluating image super-resolution algorithms. The dataset is divided into:
- Training set: 800 images for model training
- Validation set: 100 images for model validation
The images in this dataset feature diverse content including natural scenes, textures, objects, and patterns, making it one of the most challenging and realistic benchmarks for super-resolution research.
The files in this repository have been compressed using Guetzli, a JPEG encoder that optimizes for high visual quality at small file sizes. This ensures that the images retain their visual fidelity while significantly reducing their storage requirements, making the dataset more accessible for researchers and practitioners.
This dataset is commonly used for:
- Single image super-resolution (SISR)
- Image restoration tasks
- Benchmarking deep learning models
- Perceptual quality evaluation
If you use this dataset in your research, please cite the original DIV2K paper:
@inproceedings{Agustsson2017,
title={NTIRE 2017 Challenge on Single Image Super-Resolution: Dataset and Study},
author={Agustsson, Eirikur and Timofte, Radu},
booktitle={IEEE Conference on Computer Vision and Pattern Recognition Workshops},
year={2017}
}
This repository is licensed under the MIT License. See the LICENSE file for more information.