author = "Diniz, Matheus Alves and Menotti, David and Schwartz, William 
          affiliation = "{Universidade Federal de Minas Gerais} and {Universidade Federal 
                         do Paran{\'a}} and {Universidade Federal de Minas Gerais}",
                title = "Face Detection at 15,000 FPS: Real-Time Inference on GPU and CPU",
            booktitle = "Proceedings...",
                 year = "2018",
               editor = "Ross, Arun and Gastal, Eduardo S. L. and Jorge, Joaquim A. and 
                         Queiroz, Ricardo L. de and Minetto, Rodrigo and Sarkar, Sudeep and 
                         Papa, Jo{\~a}o Paulo and Oliveira, Manuel M. and Arbel{\'a}ez, 
                         Pablo and Mery, Domingo and Oliveira, Maria Cristina Ferreira de 
                         and Spina, Thiago Vallin and Mendes, Caroline Mazetto and Costa, 
                         Henrique S{\'e}rgio Gutierrez and Mejail, Marta Estela and Geus, 
                         Klaus de and Scheer, Sergio",
         organization = "Conference on Graphics, Patterns and Images, 31. (SIBGRAPI)",
            publisher = "Sociedade Brasileira de Computa{\c{c}}{\~a}o",
              address = "Porto Alegre",
             keywords = "face,detection, real-time, deep learning, cpu.",
             abstract = "Object detection is a key task in computer vision since it is the 
                         first step in the pipeline of many applications such as person 
                         re-identification, vehicle identification, and face verification. 
                         Recently, the best performing object detectors have been achieved 
                         with deep learning and one common characteristic among them is 
                         that they are a very slow on ordinary hardware. Reported real time 
                         object detectors are usually measured with high-end GPUs, which is 
                         inappropriate for scenarios where energy efficiency and low costs 
                         are required. We were able to train a very light face detection 
                         architecture by greatly reducing the number of parameters and 
                         input size of a convolutional network. Our model is capable of 
                         performing inference in real time on a hardware as simple as a 
                         Raspberry Pi. Furthermore, when evaluated on a GPU, we were able 
                         to achieve up to 15,000 frames per second.",
  conference-location = "Foz do Igua{\c{c}}u, PR, Brazil",
      conference-year = "Oct. 29 - Nov. 1, 2018",
             language = "en",
                  ibi = "8JMKD3MGPAW/3S4PPHS",
                  url = "",
           targetfile = "face_detection_at_15k_fps.pdf",
        urlaccessdate = "2020, Dec. 04"