Understanding the perceived quality of video predictions
作者：
Highlights：
• A new database - IISc Predicted Videos Quality Assessment (PVQA) database containing 300 videos suffering from various distortions due to video prediction algorithms along with corresponding human opinion scores.
• Benchmarking of popular image and video QA measures on the IISc-PVQA database - existing measures correlate poorly with human opinion.
• A new no-reference objective measure to evaluate the quality of predicted videos based on motion-compensated cosine similarity of deep features of frames and deep features of frame differences.
摘要
•A new database - IISc Predicted Videos Quality Assessment (PVQA) database containing 300 videos suffering from various distortions due to video prediction algorithms along with corresponding human opinion scores.•Benchmarking of popular image and video QA measures on the IISc-PVQA database - existing measures correlate poorly with human opinion.•A new no-reference objective measure to evaluate the quality of predicted videos based on motion-compensated cosine similarity of deep features of frames and deep features of frame differences.
论文关键词：Video quality assessment,Video prediction,Database,Perceptual quality,Neural networks,Deep learning
论文评审过程：Received 22 June 2021, Revised 14 December 2021, Accepted 23 December 2021, Available online 30 December 2021, Version of Record 17 January 2022.
论文官网地址：https://doi.org/10.1016/j.image.2021.116626
