Burke, Michael G2017-11-062017-11-062017-10Burke, M.G. 2017. Leveraging Gaussian process approximations for rapid image overlay production. SAWACMMM '17- Proceedings of the ACM Multimedia 2017 Workshop on South African Academic Participation, Mountain View, California, USA, October 23 - 23, 2017978-1-4503-5505-6https://dl.acm.org/citation.cfm?id=3132715&CFID=822088833&CFTOKEN=19372045Doi>10.1145/3132711.3132715http://hdl.handle.net/10204/9726Copyright: 2017 The Author. Paper presented at SAWACMMM '17- Proceedings of the ACM Multimedia 2017 Workshop on South African Academic Participation, Mountain View, California, USA, October 23 - 23, 2017Machine learning models trained using images can be used to generate image overlays by investigating which image areas contribute the most towards model outputs. A common approach used to accomplish this relies on blanking image regions using a sliding window and evaluating the change in model output. Unfortunately,this can be computationally expensive,as it requires numerous model evaluations. This paper shows that a Gaussian process approximation to this blanking approach produces outputs of similar quality,despite requiring significantly fewer model evaluations. This process is illustrated using a user-driven saliency generation problem. Here,pairwise image interest comparisons are used to infer underlying image interest and a Gaussian process model trained to predict the interest value of an image using image features extracted by a convolutional neural network. Interest overlays are generated by evaluating model change at blanking image regions selected using the prediction uncertainty of a Gaussian process regressor.enGaussian processesSaliency generationConference PresentationBurke, M. G. (2017). Leveraging Gaussian process approximations for rapid image overlay production. ACM Digital Library. http://hdl.handle.net/10204/9726Burke, Michael G. "Leveraging Gaussian process approximations for rapid image overlay production." (2017): http://hdl.handle.net/10204/9726Burke MG, Leveraging Gaussian process approximations for rapid image overlay production; ACM Digital Library; 2017. http://hdl.handle.net/10204/9726 .TY - Conference Presentation AU - Burke, Michael G AB - Machine learning models trained using images can be used to generate image overlays by investigating which image areas contribute the most towards model outputs. A common approach used to accomplish this relies on blanking image regions using a sliding window and evaluating the change in model output. Unfortunately,this can be computationally expensive,as it requires numerous model evaluations. This paper shows that a Gaussian process approximation to this blanking approach produces outputs of similar quality,despite requiring significantly fewer model evaluations. This process is illustrated using a user-driven saliency generation problem. Here,pairwise image interest comparisons are used to infer underlying image interest and a Gaussian process model trained to predict the interest value of an image using image features extracted by a convolutional neural network. Interest overlays are generated by evaluating model change at blanking image regions selected using the prediction uncertainty of a Gaussian process regressor. DA - 2017-10 DB - ResearchSpace DP - CSIR KW - Gaussian processes KW - Saliency generation LK - https://researchspace.csir.co.za PY - 2017 SM - 978-1-4503-5505-6 T1 - Leveraging Gaussian process approximations for rapid image overlay production TI - Leveraging Gaussian process approximations for rapid image overlay production UR - http://hdl.handle.net/10204/9726 ER -