The MediaEval 2016 Emotional Impact of Movies Task Run submissions • Up to 5 runs for each subtask • A required run which uses no external training data, only the provided development data is allowed Evaluation Metrics: • Mean Square Error • Pearson’s Correlation Coefficient Development dataset: LIRIS-ACCEDE Discrete LIRIS-ACCEDE • 9800 video clips from 160 movies under Creative Commons licenses • Duration between 8s and 12s • Cross-validated through a controlled experimental protocol Continuous LIRIS-ACCEDE • 30 movies • Duration between 117s and 4,566s (total duration: ~7 hours) • Continuous induced valence and arousal self-assessments Test dataset: • From 49 new movies under Creative Commons licenses • 1,200 additional short video clips for the first subtask (between 8 and 12 seconds) • 10 additional long movies (from 25 minutes to 1 hour and 35 minutes) for the second subtask (for a total duration of 11.48 hours) Sqdf sdf Ground truth Valence and arousal ranking: • Pairwise video comparisons on CrowdFlower • Annotators asked to focus on the emotion they felt • Simple task: • Which one conveys the most positive emotion? • Which one conveys the calmest emotion? From rankings to ratings: • Ratings collected for 40 video clips regularly distributed • 28 participants • Ratings estimated using Gaussian Process models Continuous annotation: • Induced valence and arousal self-assessments • 16 participants • Modified Gtrace interface and joystick Task Description • Deploy multimedia features and models to automatically predict the emotional impact of movies • Emotion considered in terms of induced valence and arousal Two subtasks: • Global emotion prediction: given a short video clip (around 10 seconds), participants’ systems are expected to predict a score of induced valence (negative-positive) and induced arousal (calm-excited) for the whole clip; • Continuous emotion prediction: as an emotion felt during a scene may be influenced by the emotions felt during the previous ones, the purpose here is to consider longer videos, and to predict the valence and arousal continuously along the video. Thus, a score of induced valence and arousal should be provided for each 1s- segment of the video. Context • An evolution of previous years tasks on violence and affect prediction from videos • Applications: • Personalized content delivery • Movie recommendation • Video editing supervision • Video summarization • Protection of children from potential harmful content Organizers: Emmanuel Dellandréa, Liming Chen, Yoann Baveye, Mats Sjöberg, Christel Chamaret Contact: Emmanuel Dellandréa – [email protected] Representation of emotions Credits and license information is available here: http://liris-accede.ec-lyon.fr/database.php Arousal Valence