Person Recognition in Personal Photo Collections

TL;DR

This paper introduces a convnet-based system for recognizing persons in personal photos, analyzing body cues, training data impact, and proposing more challenging benchmarks to improve accuracy in social media photo recognition.

Contribution

It presents a simple, open-source convnet approach with comprehensive analysis and new benchmarks for improved person recognition in social media images.

Findings

Improved state-of-the-art accuracy on PIPA dataset

Analysis of body cues and training data effects

Identification of common failure modes

Abstract

Recognising persons in everyday photos presents major challenges (occluded faces, different clothing, locations, etc.) for machine vision. We propose a convnet based person recognition system on which we provide an in-depth analysis of informativeness of different body cues, impact of training data, and the common failure modes of the system. In addition, we discuss the limitations of existing benchmarks and propose more challenging ones. Our method is simple and is built on open source and open data, yet it improves the state of the art results on a large dataset of social media photos (PIPA).