Places: A 10 Million Image Database for Scene Recognition

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.

Learning Context Flexible Attention Model for Long-Term Visual Place Recognition

“… FABMAP [1]: This is a state-of-the-art place recognition algorithm built on top of handcrafted features  SeqSLAM [14]: This is a sequence-based place recognition approach, which have demonstrated state-of-the-art performances on mapping environments across seasons, weather conditions and different times of a day.  Places365 [22]: This is a CNN-based scene recognition model trained to recognize 365 scene types. The model was trained on over two million pictures.…”

“…Rather than recognizing specific places, networks have also been trained for recognizing the types of places [22]. However, this scene recognition task is different in nature from visual place recognition: image under the same scene category can come from different places.…”

Quantifying Urban Surroundings Using Deep Learning Techniques: A New Proposal

“…The Places dataset [44] consists Urban Sci. 2018, 2, 78 7 of 19 of around 10 million images depicting indoor and outdoor views.…”

Predicting eye movements from deep neural network activity decoded from fMRI responses to natural scenes

“…As the foundation for our spatial attention model, we used a goal-directed CNN with a deep architecture 10 trained for scene categorization 18,19 . This architecture consists of 18 spatiallyselective layers that compute alternating convolution and non-linear max-pooling operations (Fig.…”

“…To facilitate straightforward interpretation of which factors drive significant prediction of eye movements from spatial priority maps reconstructed from fMRI activity, these steps were excluded from our model. 10,18,19 . (b) Unit activity was extracted from the five pooling layers to sample activity from across the CNN hierarchy.…”