Research-only dataset. I do not own any of the images. DF2K is a convenience merge of the public DIV2K and Flickr2K datasets commonly used for single-image super-resolution research. This mirror is provided solely to facilitate open research. See Licensing & Permissions and Takedown below.

If you are a rights holder and want any content removed, please see the Takedown section and I will promptly delete the images, portions of the dataset, or the entire dataset.

Dataset Summary

Name: DF2K
Sources: DIV2K + Flickr2K
Task: Single Image Super-Resolution (SISR) and related low-level vision tasks
Intended use: Non-commercial research and benchmarking only
Size: 3550 images for each scale (HR, LRx2, LRx3, LRx4)

This repo simply combines the original DIV2K and Flickr2K images as commonly referenced in SISR literature. No additional annotations or labels beyond what the original sources provide.

Folder structure:

DF2K-bicubic/
├── hr/   # High-resolution ground-truth images (originals from DIV2K + Flickr2K). Used as targets.
├── x2/   # Low-resolution images downsampled from HR by a factor of 2 using bicubic. For ×2 SR.
├── x3/   # Low-resolution images downsampled from HR by a factor of 3 using bicubic. For ×3 SR.
└── x4/   # Low-resolution images downsampled from HR by a factor of 4 using bicubic. For ×4 SR.

How to Use

In terminal:

pip install -U huggingface_hub git-lfs

huggingface-cli download \
  --repo-type dataset bezdarnost/DF2K-bicubic \
  --local-dir /path/to/DF2K-bicubic \
  --local-dir-use-symlinks False

Or with python:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="bezdarnost/DF2K-bicubic",
    repo_type="dataset",
    local_dir="/path/to/DF2K-bicubic",
    local_dir_use_symlinks=False,
)

Supported Tasks & Benchmarks

Single Image Super-Resolution (SISR)
Image restoration and low-level vision pretraining

Data Licenses, Ownership, and Permissions (Important)

I am not the creator or owner of DIV2K or Flickr2K.
This repository is a research mirror to make experiments more accessible.
No warranties. No commercial use.
For copyright, license, and usage terms, please refer to the original sources:
- DIV2K: https://data.vision.ee.ethz.ch/cvl/DIV2K/
- Flickr2K (NTIRE2017): https://github.com/limbee/NTIRE2017

If the original licenses or terms prohibit mirroring or require removal, I will comply—please contact me (see Contact).

Ethical Considerations

Images originate from third-party sources; please respect privacy, attribution, and any usage restrictions from the original datasets.
Do not use this dataset for applications that may infringe on privacy, cause harm, or violate original terms.

Citations

DIV2K

Collected from: https://data.vision.ee.ethz.ch/cvl/DIV2K/

@InProceedings{Agustsson_2017_CVPR_Workshops,
  author    = {Agustsson, Eirikur and Timofte, Radu},
  title     = {NTIRE 2017 Challenge on Single Image Super-Resolution: Dataset and Study},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  month     = {July},
  year      = {2017}
}

Flickr2K (NTIRE2017)

Collected from: https://github.com/limbee/NTIRE2017

@InProceedings{Timofte_2017_CVPR_Workshops,
  author    = {Timofte, Radu and Agustsson, Eirikur and Van Gool, Luc and Yang, Ming-Hsuan and Zhang, Lei and others},
  title     = {NTIRE 2017 Challenge on Single Image Super-Resolution: Methods and Results},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  month     = {July}
}

Takedown & Contact

I do not claim any rights to the images. If you are a rights holder or dataset maintainer and wish for any content to be removed:

Open an Issue on this repository or
Send an email to amanurumbekov@gmail.com with:
- A link to this dataset card
- The specific files/paths in question
- Proof of ownership or an authoritative request

I will remove the specified content promptly.

Disclaimer

This dataset mirror is provided as-is for research purposes only. I receive no compensation from hosting it. By using this dataset, you agree to comply with the terms of the original sources and all applicable laws and licenses. If any term conflicts with the original datasets’ licenses or policies, the original terms prevail.

Downloads last month: 492

DF2K (DIV2K + Flickr2K)