Subjective video quality evaluation for user generated contents via textual prompts

Abstract

The explosion of user-generated content (UGC) videos has transformed the online landscape. Platforms like TikTok and YouTube overflow with UGC, offering a diverse range of content consumed by billions of viewers. While this presents exciting avenues for creative expression and content consumption, it also introduces significant challenges in assessing video quality.

Traditionally, video quality evaluation has relied heavily on technical metrics like bitrate and resolution. However, this approach often neglects the subjective experience of viewers. Technical aspects, while crucial, do not fully capture the nuances that influence how viewers perceive the quality of a video.

This research aims to bridge this gap by acknowledging the limitations of solely relying on technical parameters. We propose a novel approach that disentangles the influence of technical quality (sharpness, focus, noise) from aesthetic quality (content, composition, color, lighting) on viewers' perception. By understanding this interplay, we can gain a deeper comprehension of how subjective quality is experienced.

Furthermore, this research goes beyond simply identifying the factors that influence perception. We aim to develop a system that translates these factors into human-understandable rationales. Current methods often present a single quality score, leaving users in the dark about the specific elements driving their experience. Our objective is to provide clear explanations that illuminate the technical and aesthetic aspects contributing to the perceived quality of a UGC video.

By achieving these goals, this research aspires to significantly enhance our understanding of UGC video quality perception. This newfound knowledge paves the way for the development of more comprehensive quality assessment methods. These methods will not only consider technical parameters but also incorporate the subjective influence of aesthetic qualities, ultimately leading to a more holistic evaluation of UGC video quality.