VRI-GazeNet: 3D Gaze Estimation for Real-time Applications

  • Gabriel N. Hishida UFPR
  • Eduardo Todt UFPR


Recognizing where humans are looking is a relevant and challenging task. Modern available AI can compute gaze direction with less than 10° of average angular error, however, these models have an inference time which are too long to be used in real-time applications such as social robots and driving monitoring systems that need to calculate gaze direction quickly. This paper introduces VRI-GazeNet, a gaze estimation network that, when compared to the state-of-the-art models, presents similar accuracy with a speedup of 2.5. Our model is available at https://github.com/VRI-UFPR/GazeNet.

Como Citar

Selecione um Formato
HISHIDA, Gabriel N.; TODT, Eduardo. VRI-GazeNet: 3D Gaze Estimation for Real-time Applications. In: SIMPÓSIO BRASILEIRO DE ROBÓTICA E SIMPÓSIO LATINO AMERICANO DE ROBÓTICA (SBR/LARS), 15. , 2023, Salvador/BA. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2023 . p. 7-10.