RFV-LQ
收藏资源简介:
RFV-LQ数据集是由香港大学创建的低质量人脸视频基准,旨在评估和扩展盲人脸图像恢复算法在视频场景中的应用。该数据集包含329条从各种真实世界视频源(如旧的脱口秀、电视剧和电影)中精心挑选的低质量人脸视频。数据集的创建过程包括使用RetinaNet进行人脸检测,SORT进行跟踪,以及ArcFace进行身份验证。RFV-LQ数据集主要用于解决视频中人脸恢复的挑战,特别是在面部细节恢复和时间一致性方面。
The RFV-LQ dataset is a low-quality face video benchmark developed by The University of Hong Kong, aiming to evaluate and extend the application of blind face image restoration algorithms in video scenarios. This dataset contains 329 low-quality face videos carefully curated from diverse real-world video sources, including vintage talk shows, TV dramas, and feature films. The dataset construction pipeline utilizes RetinaNet for face detection, SORT for face tracking, and ArcFace for identity verification. The RFV-LQ dataset is primarily designed to address the challenges of face restoration in videos, particularly in terms of facial detail recovery and temporal consistency.

- 1Analysis and Benchmarking of Extending Blind Face Image Restoration to Videos香港大学 · 2024年



