YouTube UGC Dataset for Video Compression Research

Abstract

Non-professional video, commonly known as User Generated Content (UGC) has become very popular in todays video sharing applications. However, traditional metrics used in compression and quality assessment, like BD-Rate and PSNR, are designed for pristine originals. Thus, their accuracy drops significantly when being applied on non-pristine originals (the majority of UGC). Understanding difficulties for compression and quality assessment in the scenario of UGC is important, but there are few public UGC datasets available for research. This paper introduces a large scale UGC dataset (1500 20 sec video clips) sampled from millions of YouTube videos. The dataset covers popular categories like Gaming, Sports, and new features like High Dynamic Range (HDR). Besides a novel sampling method based on features extracted from encoding, challenges for UGC compression and quality evaluation are also discussed. Shortcomings of traditional reference-based metrics on UGC are addressed. We demonstrate a promising way to evaluate UGC quality by no-reference objective quality metrics, and evaluate the current dataset with three no-reference metrics (Noise, Banding, and SLEEQ).

Advances in Video Compression System Using Deep Neural Network: A Review and Case Studies

“…We selected sequences with texture regions from standard test sequences and the more challenging YouTube UGC data set 4 [198]. The YouTube UGC data set is a sample selected from thousands of user-generated content (UGC) videos uploaded to YouTube.…”

“…In many instances, this mask drives the encoder to perform separately for pInSIG textures that are typically inferred without additional residuals and "pSIG" areas elsewhere using the traditional hybrid coding method. This approach is implemented on top of the AV1 codec [195]- [197] by enabling the GoP-level switchable mechanism, resulting in noticeable bit rate savings for both standard test sequences and additional challenging sequences from the YouTube UGC data set [198], under similar perceptual quality. The method that we propose is a pioneering work that integrates learningbased texture analysis and reconstruction approaches with modern video codec to enhance video compression performance.…”

Escaping the complexity-bitrate-quality barriers of video encoders via deep perceptual optimization

“…We report on tests with XIPH and CDVL sequences ‡ (representing prime content) and YouTube UGC 1080p 'LiveMusic' and 'Sports' (as examples of user-generated content 31 aomenc --passes=2 --pass=1 --fpf=aom1.log --target-bitrate=B --cpu-used=Pav1 --threads=8 --tile-columns=1 --tile-rows=0 --kf-max-dist=150 -o "video out.webm" "video in.y4m" 2>1 aomenc --passes=2 --pass=2 --fpf=aom1.log --target-bitrate=B --cpu-used=Pav1 --threads=8 --tile-columns=1 --tile-rows=0 --kf-max-dist=150 -o "video out.webm" "video in.y4m" 2>1…”

“…BD-rate results for DPO enh0m+enh3m models and H.264/AVC. The sequences are from YouTube UGC dataset 31. Videos marked with an asterisk (*) are excluded from the averages and are discussed separately.…”

Testing self-adaptive software with probabilistic guarantees on performance metrics

“…According to the proposed methodology, we define a set of videos that can be considered a random sample, with respect to their content. Here, we used the User Generated Content dataset from Youtube [67]. This dataset is representative of videos uploaded by users to Youtube.…”