KAIST IVY LAB Omnidirectional Image Database for Visual Quality Assessment

For training of our model and performance evaluation, we utilized SUN360 database, which is a large scale of 360 degree images represented in equirectangular projection with 9104 x 4552 pixels. In our subjective VR experiment, A total of 15 subjects participated. Oculus Rift CV1 with the Oculus 360 Photos was used to display the omnidirectional images. All subjects were seated on a rotatable chair. All experimental settings followed the guideline, ITU-R BT.500-13 and BT.2021.
This web page provides the datasets of our ICASSP 2018 paper [1], so that researchers can repeat our experiments or test their VR-IQA methods on our datasets. The dataset is for research purposes only. If you use our datasets, please cite the paper [1].

Omnidirectional Image Datasets

The datasets provided below are image datasets from SUN360 database and visual quality score obtained by subjective assessment experiment. The images are 60 randomly selected omnidirectional images in SUN360. In experiment, in order to match the resolution of our HMD, the original high resolution omnidirectional images were down-sampled using bi-cubic interpolation to 2048 x 1024 pixels. To generate distorted images, we compressed the 60 omnidirectional images using three widely used codec standards, JPEG, JPEG 2000, and HEVC. In experiment, we selected four different bit rates, 0.5, 1.0, 1.5 and 2.0 bits per pixel for each compression. The number of total omnidirectional images obtained is 720 (60 scenes x 4 codec standards x 4 bit rates).

Omnidirectional images selected from SUN360 and distorted images: [Link]

Corresponding Subjective Image Quality Score Datasets

The visual quality score is measured using single stimulus continuous quality evaluation (SSCQE). The number of subjects participated in subjective VR experiment is 15. The subjects scored their perceived quality in the continuous scale range of 0-100, divided into five grades: excellent, good, fair, poor, and bad.

Image quality score obtained subjective assessment experiment : [Link]

Reference

If you use the database, please cite as :

[1] Lim, H.T., Kim, H.G., & Ro Y.M. (2018) VR IQA NET: Deep Virtual Reality Image Quality Assessment using Adversarial Learning. 6737-6741. 10.1109/ICASSP.2018.8461317.

ⓒ2018 IMAGE & Video SYSTEMS LAB DB.
KOREA ADVANCED IN STITUTE OF SCIENCE AND TECHNOLOGY(KAIST) Tel. +82-42-350-5494, 8094 335 Gwahak-ro(373-1 Guseong-dong). Yuseong-gu, Daejeon 305-701,Republic of Korea