CVQAD (Compressed Video Quality Assessment Dataset)
1,022 compressed videos (including UGC), 32 encoders of 5 compression standards (H.265/HEVC, H.266/VVC, AV1, H.264/AVC, and VP9), 3 target bitrates (1,000 kbps, 2,000 kbps, and 4,000 kbps).
Description
The CVQA dataset focuses on the compressed video quality assessment problem. It contains 1,022 distorted streams, produced from 36 reference videos, which were originally collected using space-time-complexity clustering from a pool of more that 18,000 videos with the average bitrate of 130 Mbps. Coding artifacts were obtained by compressing videos through several encoders (32 in total) of different compression standards: H.265/HEVC, H.266/VVC, AV1, H.264/AVC, and VP9. There were two different presets (30 FPS and 1 FPS) and three target bitrates — 1,000 kbps, 2,000 kbps, and 4,000 kbps. More than 320,000 subjective scores were collected through the Subjectify.us crowdsourcing platform, which employs a Bradley-Terry model to transform the results of pairwise voting into a score for each video. Content includes nature, sport, humans close up, gameplays, music videos, water stream or steam, CGI, and user-generated content. Several effects and distortions are also presented: shaking, slow-motion, grain / noisy, too dark / bright regions, macro shooting, captions (text), and extraneous objects.
Access
Fill out the form on the CVQAD page, and the dataset will be sent to you within a few hours.
License
CCBY
References and Citation
If you use this database in your research, we kindly ask that you follow the copyright notice and cite the following paper (you can also find the Bibtex citation template on the paper page)
References
ANT22: Anastasia Antsiferova, Sergey Lavrushkin, Maksim Smirnov, Aleksandr Gushchin, Dmitriy Vatolin, and Dmitriy Kulikov, “Video compression dataset and benchmark of learning-based video-quality metrics,” in Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2022.