YouTube SFV+HDR Quality Dataset

Overview

• The popularity of Short-form Videos (SFV) has grown dramatically in recent years, with billions of viewers.
• High Dynamic Range (HDR) has also become more popular on video sharing platforms.
• This paper explores whether SFV+HDR quality assessment is significantly different from traditional User Generated Content (UGC) quality assessment, and whether existing objective quality metrics designed for UGC still work well for SFV+HDR.

Plain English Explanation

The researchers created the first large-scale dataset of Short-form Videos (SFVs) with High Dynamic Range (HDR), which includes subjective quality scores. This dataset covers 10 popular content categories and uses a sampling framework to ensure it is representative.

The researchers then analyzed the subjective quality scores for short SDR (Standard Dynamic Range) and HDR videos. They also evaluated how well the state-of-the-art UGC (User Generated Content) quality metrics perform on this new SFV+HDR dataset, and discussed potential improvements.

The key questions the researchers aimed to answer are:

1. Is the quality assessment for SFV+HDR significantly different from traditional UGC quality assessment?
2. Do the objective quality metrics designed for traditional UGC still work well for SFV+HDR?

Technical Explanation

The researchers created a new large-scale dataset of SFV+HDR videos, with reliable subjective quality scores, to investigate these questions. They used a sampling framework to maximize the representativeness of the dataset, covering 10 popular content categories.

The researchers then conducted a comprehensive analysis of the subjective quality scores for the short SDR and HDR videos in the dataset. They also evaluated the performance of state-of-the-art UGC quality metrics on this new SFV+HDR dataset, and discussed potential areas for improvement.

Critical Analysis

The researchers acknowledge that their dataset is the first of its kind, covering SFV+HDR content. This is an important step forward in video quality research, as the growing popularity of these video formats necessitates a better understanding of how to assess their quality.

However, the researchers do not discuss the potential limitations of their dataset or sampling framework. It would be helpful to understand if there are any biases or gaps in the content coverage, and how that could impact the generalizability of their findings.

Additionally, the paper does not provide a detailed analysis of why the existing UGC quality metrics may or may not perform well on the SFV+HDR dataset. Further insights into the specific factors that contribute to quality differences between traditional UGC and SFV+HDR would strengthen the paper's conclusions.

Conclusion

This research highlights the need for a deeper understanding of video quality assessment for emerging video formats, such as Short-form Videos (SFVs) with High Dynamic Range (HDR). The new dataset and analysis presented in this paper provide a valuable foundation for future work in this area.

By exploring the differences between SFV+HDR and traditional UGC quality assessment, as well as the performance of existing quality metrics, the researchers have identified an important research direction. Continued efforts to develop robust and reliable quality assessment methods for these new video formats will be crucial as they continue to gain popularity.