Detecting Violent Content in Hollywood Movies and User-Generated Videos - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 300

Table 11.2 The Mean Average Precision (MAP) and MAP@100 of our method with different
video representations (i.e., audio-only where we use MFCC-based mid-level audio representations,
visual-only where HoG- and HoF-based mid-level features and ViF descriptors are used, affect-only
where we use affect-related color features, and multimodal where the previous three representations
are linearly fused at the decision level) on the Hollywood movie dataset
Dataset

MAP MAP@100

Audio-only 0.363 0.476

Visual-only 0.327 0.439

Affect-only 0.209 0.140

Multimodal 0.422 0.539

Table 11.3 The Mean Average Precision (MAP) and MAP@100 of our method with different
video representations (i.e., audio-only where we use MFCC-based mid-level audio representations,
visual-only where HoG- and HoF-based mid-level features and ViF descriptors are used, affect-only
where we use affect-related color features, and multimodal where the previous three representations
are linearly fused at the decision level) on the Web video dataset
Dataset MAP MAP@100

Audio-only 0.582 0.582

Visual-only 0.478 0.478

Affect-only 0.495 0.495

Multimodal 0.567 0.567

features does not help improving the performance in the case of short web videos.
This might be the consequence of the weights used in the fusion process, as more
weight is given to visual-only representations. On the contrary, affect-related features
perform better in web videos.
In order to allow the interested reader to have a closer look at the evaluation results,
we provide movie-specific MAP@100 values of our best-performing method which
is the one with the multimodal representation (Table 11.4 ). One significant point

Table 11.4 The movie-specificMAP@100 values of our best-performingmethod (i.e., the one with
a multimodal representation) on the Hollywood movie dataset. Each video segment is of length 0.6 s
Movie name MAP@100 # of Violent video segments

Fantastic Four 1 0.615 2,102

Fargo 0.416 1,426

Forrest Gump 0.548 1,139

Legally Blond 0.000 0

Pulp Fiction 0.992 3,720

The God Father 1 0.290 987

The Pianist 0.910 2,220