Perceptual Quality Assessment of Omnidirectional Images

This repository contains the constructed omnidirectional image quality assessment database and implementations for the paper "Perceptual Quality Assessment of Omnidirectional Images", Yuming Fang, Liping Huang, Jiebin Yan, Xuelin Liu, and Yang Liu, Thirty-Sixth AAAI Conference on Artificial Intelligence, 2022.

VR IQA Database

The proposed omnidirectional image quality assessment database and the annotations (MOS, HMD data) can be downloaded at the Baiduyun (Password: 92rc) or Google drive.

  • Catalog:
    ├─ AAAI2022
    │        ├─ ref     // reference omnidirectional images, 1.png~258.png
    │        ├─ dis     // distorted omnidirectional images, 1_len3_bd_3.png
    │        ├─ HMData.zip     // head/eye movement data
    │        │         ├─ bad    // group
    │        │         │      ├─ A    // session
    │        │         │      ├─ B    // session
    │        │         │      ├─ C    // session
    │        │         │      └─ D    // session
    │        │         └─ good    // group
    │        │                 ├─ A    // session
    │        │                 ├─ B    // session
    │        │                 ├─ C    // session
    │        │                 └─ D    // session
    │        └─ mos.xls     // mean opinion scores
    │        └─ startpoint.xls     // startpoints
  • Explanation:
      1. ref: 258 reference omnidirectional images
      2. dis: 1032 distorted omnidirectional images
             — Nameing rule: No.reference_No.lens_Dist-type_No.level.png
             — Distortion type: bd: brightness discontinuity, gb: gaussian blur, gn: gaussian noise, st: stitching
      3. HMData: head/eye movement data of subjects, including two groups(bad/good start point) and four sessions in each group, each session contains data from 15 subjects. Revised version was uploaded on Oct 8, 2022.
             — bad
                   — A: 1,2,...,15
                   — B: 1,2,...,15
                   — C: 1,2,...,15
                   — D: 1,2,...,15
             — good
                   — A: 1,2,...,15
                   — B: 1,2,...,15
                   — C: 1,2,...,15
                   — D: 1,2,...,15
        Data format: csv
             — Head movement data
                   A col: pitch
                   B col: yaw
                   C col: roll
             — Eye movement data
                   D col: latitude
                   E col: longitude
                   F col: the eye fixation is valid (1) or invalid (0)
      4. mos.xls: mean opinion scores
      5. startpoint.xls: startpoints
sessionimgNameaveragestddistortionTypelevelexplorationTimestartingPointScene
C_good1_len3_bd_3.png3.5330.499BD35s1CreativePark
···························

Proposed Model

TBC.

Citation

Please cite our papers if it helps your research:

@inproceedings{fang2022aaai,
title={Perceptual Quality Assessment of Omnidirectional Images},
author={Fang, Yuming and Huang, Liping and Yan, Jiebin and Liu, Xuelin and Liu, Yang},
booktitle={Thirty-Sixth AAAI Conference on Artificial Intelligence},
pages={580-588},
year={2022}
}
@article{liu2024tomm,
title={Perceptual Quality Assessment of Omnidirectional Images: A Benchmark and Computational Model},
author={Liu, Xuelin and Yan, Jiebin and Huang, Liping and Fang, Yuming and Wan, Zheng and Liu, Yang},
journal={ACM Transactions on Multimedia Computing, Communications and Applications},
year={2024}
}