KOKA-10K dataset

Dataset Description & Download InstructionsThis dataset contains approximately 10,000 labeled images and ~140,000 pristine unlabeled images.The labeled subset is sourced from KONIQ-10K, while the unlabeled images originate from the KADIS-700K collection.
Due to the 20 GB upload limitation on Figshare, we are unable to include the unlabeled dataset directly in this repository. Instead, the unlabeled set (≈ 45.5 GiB) is provided externally through a multi-part download.
External Links to Original Data Sources

KONIQ-10K (Labeled data)
https://database.mmsp-kn.de/koniq-10k-database.html
KADIS-700K (Source of pristine unlabeled data)
https://database.mmsp-kn.de/kadid-10k-database.html
Unlabeled Dataset (≈ 45.5 GiB)

The unlabeled dataset is divided into 12 compressed parts due to hosting constraints.
You must download all 12 parts before extraction.
Download link: https://terabox.com/s/1l-g5CQzMcYUl7Bz3KYbaDA
Each part is named sequentially, e.g: unlabel-part_aa, unlabel-part_ab, ... unlabel-part_al
After all parts have been downloaded into the same directory, merge them back into the original compressed file using by command below:
Linux/MacOS: cat unlabel-part_* > unlabel.zip
Window (Powershell): Get-Content unlabel-part_* -Encoding Byte -ReadCount 0 | Set-Content unlabel.zip -Encoding Byte