metadata

license: apache-2.0
task_categories:
  - image-to-text
language:
  - en
tags:
  - multi-modal image quality assessment
pretty_name: Data-DeQA-Score
size_categories:
  - 10K<n<100K

Data-DeQA-Score

Datasets of the DeQA-Score paper ( project page / codes / paper ) in our DepictQA project.

Dataset Construction

Download our meta files in this repo.
Download source images from KonIQ, SPAQ, KADID, PIPAL, LIVE-Wild, AGIQA, TID2013, and CSIQ.
Arrange the folders as follows:

|-- Data-DeQA-Score
  |-- KONIQ
    |-- images/*.jpg
    |-- metas
  |-- SPAQ
    |-- images/*.jpg
    |-- metas
  |-- KADID10K
    |-- images/*.png
    |-- metas
  |-- PIPAL
    |-- images/Distortion_*/*.bmp
    |-- metas
  |-- LIVE-WILD
    |-- images/*.bmp
    |-- metas
  |-- AGIQA3K
    |-- images/*.jpg
    |-- metas
  |-- TID2013
    |-- images/distorted_images/*.bmp
    |-- metas
  |-- CSIQ
    |-- images/dst_imgs/*/*.png
    |-- metas

If you find our work useful for your research and applications, please cite using the BibTeX:

@inproceedings{deqa_score,
  title={Teaching Large Language Models to Regress Accurate Image Quality Scores using Score Distribution},
  author={You, Zhiyuan and Cai, Xin and Gu, Jinjin and Xue, Tianfan and Dong, Chao},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
  year={2025},
}