GS-Blur
收藏资源简介:
GS-Blur数据集是由首尔国立大学创建的一个用于真实图像去模糊的3D场景数据集。该数据集通过3D高斯喷射(3DGS)技术从多视角图像中重建3D场景,并沿随机生成的运动轨迹渲染模糊图像,从而生成大量真实且多样化的模糊图像。数据集包含156209条数据,涵盖了多种模糊类型和轨迹,旨在解决现有数据集在模糊多样性和真实性方面的不足。GS-Blur数据集的应用领域主要集中在图像去模糊技术,旨在提高去模糊网络在真实世界模糊图像上的泛化能力。
GS-Blur is a 3D scene dataset for real-world image deblurring created by Seoul National University. This dataset reconstructs 3D scenes from multi-view images using 3D Gaussian Splatting (3DGS) technology, and renders blurred images along randomly generated motion trajectories to generate a large number of realistic and diverse blurred samples. The dataset contains 156,209 entries, covering various blur types and motion trajectories, aiming to address the shortcomings of existing datasets in terms of blur diversity and realism. The main application field of the GS-Blur dataset is image deblurring technology, which aims to improve the generalization ability of deblurring networks on real-world blurred images.

- 1GS-Blur: A 3D Scene-Based Dataset for Realistic Image Deblurring首尔国立大学 · 2024年



