MediaEval 2016 - Simula Team @ Context of Experience Task
0 likes185 views
The document discusses a study conducted by the Simula team on data preparation and classification of trailers using metadata and visual features. It outlines the use of the WEKA implementation and evaluates the effectiveness of different classification runs based on metadata, visual features, and their combination. The findings suggest that global features perform best, highlighting the need for further research on feature fusion and data collection from actual travelers.
1 of 8
Download to read offline
More Related Content
Viewers also liked (12)
Ad
Similar to MediaEval 2016 - Simula Team @ Context of Experience Task (20)
Ad
More from multimediaeval (20)
Recently uploaded (20)
MediaEval 2016 - Simula Team @ Context of Experience Task
- 1. Simula team MediaEval 2016 Context of Experience Task Konstantin Pogorelov, Michael Riegler, Pål Halvorsen, Carsten Griwodz Simula Research Laboratory University of Oslo Norway konstantin, michael, paalh, griff@simula.no
- 2. Data preparation • Trailers (links provided) • Metadata – Rating – Country – Language – Year – Genre – Runtime – Rotten Tomatoes score – IMDB score – Metacritic score Numeric values Visual features Missing values filled manually or replaced by average value
- 3. Features extraction Trailers Sets of frames Visual features vectors Global features The LIRE web page: http://www.lire-‐project.net/ 21 global features used: JCD, CEDD, PHOG, color layout, etc.
- 4. Classification We used WEKA implementation of PART algorithm with default settings Metadata / Features WEKA PART + -‐ -‐ + + Good? Metadata / Features WEKA PART -‐ -‐ + + Good? + P(+) P(-‐) Train: Test:
- 5. Three runs 1: Only metadata 2: Only visual features 3: Combined – Find classes probabilities using visual features – Add classes probabilities as new metadata attributes – Classify using extended metadata vectors
- 6. Evaluation results Run Precision Recall F1-‐score TP FP TN FN Metadata 0.6047 0.9338 0.7341 127 83 4 9 Visual features 0.6333 0.9779 0.7687 133 77 10 3 Combined 0.6084 0.7426 0.6688 101 65 22 35 Baseline 0.629 0.5735 0.6 78 46 41 58
- 7. Conclusion • Global features are the best • Which features and metadata attributes are really important? • More research needed for efficient features fusion • More data! • Collect data from actual travelers