MediaEval 2018: Multimedia Evaluation Tasks
Regular Registration Deadline: 30 June
MediaEval (Benchmarking Initiative for Multimedia Evaluation) offers
shared-tasks to the multimedia research community involving images,
text, video, and audio/music. The tasks address cutting-edge multimedia
challenges (multimedia retrieval, analysis, access and exploration) with
a clear human or social aspect. Our larger aim is to promote
reproducible research that makes multimedia a positive force for society.
Further information and the link for registration to participate in the
tasks is available at: www.multimediaeval.org/mediaeval2018
Short descriptions of the tasks are below.
The first tasks start to release data in the last week of May, so we
advise you to register as soon as possible. Regular registration will
remain open through 30 June. Final data releases are in July.
See data release deadlines for the individual tasks at
Task results will be presented at the MediaEval 2018 Workshop, 29-31
October 2018 near Nice, France. The workshop requires a separate
registration. More information on the workshop will be posted to
www.multimediaeval.org during the course of the summer.
MediaEval 2018 Tasks:
Multimedia Satellite Task: Emergency Response for Flooding Events
Classify social images (Twitter) and/or detect floods in satellite
images. Explore the combination.
Medico: Multimedia for Medicine
Detect diseases in images and videos of the human gastrointestinal
tract. Minimize the required resources. Explore report generation.
AcousticBrainz Genre Task: Content-based music genre recognition from
Create a system that goes from features to genre labels. Tackle a
large-scale collection. Explore the subjectivity of genre.
Emotional Impact of Movies
Predict induced valence and arousal scores for movies (continuous
scores) and/or identify segments of movie inducing fear in kids.
Predicting Media Memorability
Create a system that goes from video segments to memorability scores.
Exploit visual features from the content or textual features from titles.
Human Behavior Analysis: No-Audio Multimodal Speech Detection in Crowded
Leverage video and wearable sensor data for speech/non-speech detection
from body movement.
GameStory: Video Game Analytics Challenge
Create an engaging and captivating summary in the form of a story that
boils down the thrill of a game to its mere essence. Offered in
collaboration with ZNIPE.TV, a rapidly growing platform for e-sport
Recommending Movies Using Content: Which content is key?
Use trailers and movie clips to predict the average ratings assigned to
movies by a user, and the variance of these ratings. Discover which
aspects of movies are most promising to improve recommendation.
Enhance images to prevent content-based analysis from inferring location
information, while at the same time maintaining or increasing their
appeal to users.
NewsREEL Multimedia: News recommendation with image/text content
Predict popular news items with image and textual analysis.
For more information on the mission of MediaEval check out the videos
and proceedings from previous workshops, e.g.,
If you have further questions, please contact: Martha Larson
[log in to unmask]
Martha Larson -- [log in to unmask]
Multimedia Computing Group
Delft University of Technology
Mekelweg 4, 2628 CD DELFT
office tel: +31(0)15 278 7357