Paper Image

Learning to assess video quality from natural data

Published on:

1 August 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Hongbo Liu,

Mingda Wu,

Kun Yuan,

Ming Sun,

Yansong Tang,

Chuanchuan Zheng,

Xing Wen,

Xiu Li


Key Details

Leverages diverse pretrained models containing useful quality cues

Adaptively weights model contributions per sample

Transfers knowledge to a lightweight assessor, reducing inference cost

Achieves state-of-the-art on 3 benchmarks without extra QA data

AI generated summary

Learning to assess video quality from natural data

This paper proposes a framework to assess the quality of videos by leveraging diverse pretrained models, without needing extra annotated quality data. It selects models pretrained on classification tasks using varied datasets, architectures and pretext tasks. These contain useful quality-related information on content, distortion and motion. An adaptive module weights each model's contribution per sample. This representation transfers knowledge to a lightweight assessor.

Answers from this paper


No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up