遇见数据集

IST-GPPDataset

收藏
DataCite Commons2021-12-09 更新2025-04-16 收录
官方服务:

资源简介:

The subjective assessment of the generalized perspective projection (parameterized with projection center d) was conducted using eight omnidirectional images, in equirectangular format, taken from the Salient360! dataset [1]. The spatial resolution of images 2, 6, and 8 is 7500×3750 pixels (8K), and for the remaining ones is 3840×1920 pixels (4K). The selected images have different types of contents, including indoor and outdoor scenes, objects near and far away from the camera, and the presence or absence of people. For each image, the viewports were rendered for three different viewing directions: front view, 45° to the right, and 45° to the left; for each viewing direction, ten viewports were produced, each one corresponding to a pair (d, FoV), with d∈{0,0.25,0.5,0.75,1} and "FoV"∈{90°,110°}. Higher values of d have not been considered, since the amount of distortion (fisheye effect) introduced by these projections is visually annoying to many of viewers. The viewport has a spatial resolution of 856×856 pixels.The Stimulus Comparison Adjectival Categorical Judgment (SCACJ) was selected as an evaluation method, since it is easier for subjects to select the most pleasant viewport, in a pair of viewports rendered with different projections, than to directly rate the viewports. Since the rectilinear projection (d=0) is typically used for the rendering of 360° images and videos, the resulting viewports were used as reference stimulus. More details about the subjective assessment procedure can be found in [2].[1] J. Gutiérrez, E. David, A. Coutrot, M. Perreira Da Silva, P. Le Callet, “Introducing UN Salient360! Benchmark: A platform for evaluating visual attention models for 360 contents”, International Conference on Quality of Multimedia Experience (QoMEX), Sardinia, Italy, May. 2018.[2] F. Jabar, J. Ascenso, and M. P. P.Queluz, ‘Objective Assessment of Perceived Geometric Distortions in Viewport Rendering of 360° Images’, IEEE J. Sel. Top. Signal Process., vol. 14, no. 1, pp. 49–63, Jan. 2020, doi: 10.1109/JSTSP.2019.2962970.

提供机构:
IEEE DataPort
创建时间:
2021-12-09
搜集汇总
背景与挑战
背景概述
IST-GPPDataset是一个用于评估360°图像视口渲染中几何失真主观质量的数据集,包含8张来自Salient360!数据集的等距柱状全景图。通过改变投影中心参数d和视场角生成视口,采用SCACJ方法进行主观比较,得到每个视口的CMOS分数。该数据集可用于研究和优化全景图像的投影渲染算法。
以上内容由遇见数据集搜集并总结生成
二维码
社区交流群
二维码
科研交流群
商业服务