Movie genre classification: A multi-label approach based on convolutions through time

摘要: Abstract The task of labeling movies according to their corresponding genre is a challenging classification problem, having in mind that an immaterial feature cannot be directly pinpointed any the movie frames. Hence, off-the-shelf image approaches are not capable handling this straightforward fashion. Moreover, may belong multiple genres at same time, making assignment typical multi-label which per se much more than standard single-label classification. In paper, we propose novel deep neural architecture based on convolutional networks (ConvNets) for performing movie-trailer It encapsulates ultra-deep ConvNet with residual connections, and it makes use special layer extract temporal information from image-based features prior mapping trailers genres. We compare proposed approach current state-of-the-art methods employ well-known descriptors other low-level handcrafted features. Results show our method substantially outperforms task, improving performance all