Places: A 10 Million Image Database for Scene Recognition

TLDR

The rise of multi‑million‑item dataset initiatives has enabled data‑hungry machine learning algorithms to reach near‑human semantic classification performance at tasks such as visual object and scene recognition. The paper introduces the Places Database, a 10‑million‑image repository of scene photographs labeled with semantic categories covering a wide range of environments. The authors compiled 10 million scene images, annotated them with semantic categories, and trained Convolutional Neural Networks (CNNs) to produce baseline scene classifiers. The trained Places‑CNNs outperform prior methods, reveal that object detectors emerge as intermediate representations, and together with the database provide a valuable resource for future scene‑recognition research.

Abstract

The rise of multi-million-item dataset initiatives has enabled data-hungry machine learning algorithms to reach near-human semantic classification performance at tasks such as visual object and scene recognition. Here we describe the Places Database, a repository of 10 million scene photographs, labeled with scene semantic categories, comprising a large and diverse list of the types of environments encountered in the world. Using the state-of-the-art Convolutional Neural Networks (CNNs), we provide scene classification CNNs (Places-CNNs) as baselines, that significantly outperform the previous approaches. Visualization of the CNNs trained on Places shows that object detectors emerge as an intermediate representation of scene classification. With its high-coverage and high-diversity of exemplars, the Places Database along with the Places-CNNs offer a novel resource to guide future progress on scene recognition problems.