BurstDeflicker
收藏资源简介:
BurstDeflicker是一个用于动态场景中闪烁去除的多帧图像数据集。该数据集由三个互补的数据收集策略构建:首先,使用Retinex理论开发了一种合成流程,能够生成具有多样性和真实感的闪烁图像;其次,从不同场景中捕获了4000个真实世界的闪烁图像,帮助模型更好地理解真实闪烁特征;最后,由于动态场景的非重复性,提出了绿幕方法来将运动引入图像对中,同时保持真实的闪烁退化。BurstDeflicker数据集包括无限数量的合成图像、4000个真实世界的闪烁图像对以及3690个从真实静态数据生成的绿幕动态图像对。
BurstDeflicker is a multi-frame image dataset for flicker removal in dynamic scenes. This dataset is constructed via three complementary data collection strategies: first, a synthetic generation pipeline based on the Retinex theory is developed to generate diverse and photorealistic flicker images; second, 4000 real-world flicker images are captured from diverse scenes to help the model better understand real flicker characteristics; third, owing to the non-repeatability of dynamic scenes, a green-screen method is proposed to introduce motion into image pairs while preserving authentic flicker degradation. The BurstDeflicker dataset includes an unlimited number of synthetic images, 4000 real-world flicker image pairs, and 3690 green-screen dynamic image pairs generated from real static data.

- 1通过南开大学计算机学院, 鹏城实验室, 香港理工大学, OPPO研究院 · 2025年



