Preference prediction based on a photo gallery analysis with scene recognition and object detection

In this paper user modeling task is examined by processing a gallery of photos and videos on mobile device. We propose novel engine for preference prediction based scene recognition, object detection facial analysis. At first, all faces in are clustered private with from large clusters processed the embedded system offline mode. Other may be sent to remote server analyzed very deep models. The visual features each photo obtained recognition These aggregated into single descriptor neural attention block. proposed pipeline implemented Android platform. Experimental results subset Photo Event Collection, Web Image Dataset Recognition Amazon Fashion datasets demonstrate possibility process images efficiently without significant accuracy degradation. source code application publicly available at https://github.com/HSE-asavchenko/mobile-visual-preferences.