ИСТИНА |
Войти в систему Регистрация |
|
ИСТИНА ИНХС РАН |
||
This talk presents the results of our stereoscopic quality assessment of over 140 full-length movies. We evaluated them with 18 objective metrics that detect technical problems in stereoscopic movies, including 5 unique metrics that estimate quality of 2D-to-3D conversion, for example edge-sharpness mismatch and stuck-to-background objects. As a result we published 10 reports with detailed visualizations of technical errors and contributions from 32 professional stereographers, totaling 3,300 pages. A lot of work has been performed on the problems of 2D-to-3D conversion, as stereoscopic-movie makers are moving away from 3D shooting in favor of 2D-to-3D conversion to avoid 3D-shooting-related stereoscopic artifacts. However, converted S3D-movies still encounter a range of problems that we demonstrate for recent stereoscopic movies, including “Pokémon: Detective Pikachu» (2019, $150M) and “Shazam!” (2019, $100M). Another important topic of this talk is subjective evaluation of stereoscopic video quality. It is important to know how the viewer perceives stereoscopic scenes with particular artifacts to fully understand the technical quality of analyzed movies and prioritize correction of certain scenes. To address this issue we conducted the largest study of over 300 viewers' reactions to stereoscopic errors, acquiring 22,200 discomfort scores, which we used to develop visual fatigue estimation models. Additionally, to further improve objective quality metrics and make their responses be closer to the viewer perception it is important to use visual saliency. For a pleasant viewing experience it is more important to fix problems on salient objects rather than in the corners of the frame. We have collected a stereoscopic saliency dataset with 50 observers. The end goal of our research is a universal stereoscopic quality metric that predicts the percentage of viewers who will experience discomfort after watching the movie.