Automatic Personality Recognition in Videos using Dynamic Networks and Rank Loss

There are a few difficulties with current automatic personality recognition technologies. Two of these are discussed in this article. They use of very brief video segments or individual frames to come to conclusion with personality factors rather than long-term behavior; and absence of techniques to record individuals' facial movements for personality recognition. To address these concerns, this work first offers a unique Rank Loss for self-regulated learning of facial movements that uses the innate time related development of facial movements in lieu of personality traits. Our method begins by training a basic U-net type system that can predict broad facial movements from a collection of unlabeled face recordings. The robust model is frozen subsequently, and a series of intermediary filters is added to the architecture. The self-regulated education is then restarted, but only with films tailored to the individual. As a result, the weights of the learnt filters are individual-specific, making it a useful tool for simulating individual facial dynamics. The weights of the learnt filters are then concatenated as an individual-specific representation, to forecast personality factors without the assistance of other components of the network. The proposed strategy is tested on ChaLearn personality dataset. We infer that the tasks performed by the individual in the video matter, merging or combined application of tasks achieves the high-rise precision. Also, multi-scale characteristics are better penetrating than single-scale dynamics, along with achieving impressive outcomes as process innovation in prediction of the personality factors scores through videos.

@article{Sathyanarayana2024,
  title     = {Automatic Personality Recognition in Videos using Dynamic Networks and Rank Loss},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {5},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Nethravathi Periyapatna Sathyanarayana and Karuna Pandith and Manjula Sanjay Koti and Rajermani Thinakaran},
  doi       = {10.14569/IJACSA.2024.0150591},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150591}
}