Movie Trailer Genre Classification Using Multimodal Pretrained Features

TL;DR

This work introduces a novel method for movie genre classification, capitalizing on a diverse set of readily accessible pretrained models that outperforms state-of-the-art movie genre classification models in terms of precision, recall, and mean average precision (mAP).

Abstract

We introduce a novel method for movie genre classification, capitalizing on a diverse set of readily accessible pretrained models. These models extract high-level features related to visual scenery, objects, characters, text, speech, music, and audio effects. To intelligently fuse these pretrained features, we train small classifier models with low time and memory requirements. Employing the transformer model, our approach utilizes all video and audio frames of movie trailers without performing any temporal pooling, efficiently exploiting the correspondence between all elements, as opposed to the fixed and low number of frames typically used by traditional methods. Our approach fuses features originating from different tasks and modalities, with different dimensionalities, different temporal lengths, and complex dependencies as opposed to current approaches. Our method outperforms state-of-the-art movie genre classification models in terms of precision, recall, and mean average precision (mAP). To foster future research, we make the pretrained features for the entire MovieNet dataset, along with our genre classification code and the trained models, publicly available.

Figures (6)

• Figure 1: Feature extraction pipeline.
• Figure 2: Multilayer perception (MLP). The last layer of the MLP block is the classification layer.
• Figure 3: Single-transformer. Regarding the transformer's output, only the output vector that corresponds to the $<$CLS$>$ vector is passed to the final classification layer. The $<$SEP$>$ vectors are used to explicitly separate the elements of each modality.
• Figure 4: Multi-transformer. Regarding the output of each transformer, only the output vectors that correspond to the $<$CLS$>$ vectors are used. For inputs that are averaged along the temporal dimensions, no $<$CLS$>$ vectors are used since temporal averaging already yields a single vector.
• Figure 5: Baseline model that works with 2-dimensional raw video frames and audio (spectrogram) frames.