Fig 1.
Influence of each descriptor, using Kolmogorov-Smirnov test.
Table 1.
Final list of descriptors.
Table 2.
Distribution of films by rating.
Fig 2.
Comparison of Histogram descriptors.
Fig 3.
Hierarchical clustering, where final nodes are the selected groups for classification.
Table 3.
Films of dataset and chosen groups.
Table 4.
Results of the classification.
Table 5.
Distribution of films of the dataset by decade.
Fig 4.
MAE of production year prediction test.
Table 6.
Production year prediction results.
Fig 5.
Results of the year prediction using a neural network.
Table 7.
Production year prediction results using an artificial neural network.
Fig 6.
Complete genre classification technique (our implementation with styles).
Table 8.
Genre classification results.
Fig 7.
Neural network architecture for genre classification using visual descriptor and image descriptors.
Fig 8.
General architecture of the ResNET NN.
In this work the last layer is removed to use a pre-trained model.
Table 9.
Genre classification details using ANN for the proposed 200 movies dataset.
Table 10.
Genre classification results, with deep learning.
Fig 9.
Films (movies) distribution according to its genre.
Genre follow the aforementioned list and were proposed in [11].
Fig 10.
Films (movies) distribution according to its year of production.
This distribution follows the aforementioned list and were proposed in [11].
Table 11.
Genre classification using Deep Learning for LMTD movies dataset.
Table 12.
Genre classification comparison of the proposal method with existing approaches, measured with the AUC (Area Under the Curve) metric as in [11].
Fig 11.
Accuracy evolution, using all relevant visual, descriptors and audio features for 50 epochs.
Fig 12.
Spearman correlation of popularity prediction.
Table 13.
Popularity prediction results.
Fig 13.
Proposed aesthetic model and possible applications.