Places: A 10 Million Image Database for Scene Recognition

By Bolei Zhou, Àgata Lapedriza, Aditya Khosla, Aude Oliva, Antonio Torralba. 2018

4,862 citations recorded by the source corpus.

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.

Subfields: Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

High-Resolution Image Synthesis with Latent Diffusion Models (2021, methodology): “We describe the exact training & evaluation protocol on Places [104] in Sec.”
Do Better ImageNet Models Transfer Better? (2018, methodology): “works pretrained on the Places dataset [29, 77], which more closely matches the domain of SUN397.”
Exploring the Limits of Weakly Supervised Pretraining (2018, methodology): “We evaluate Instagram pretraining by measuring classification accuracies on three classification target tasks: ImageNet [30], CUB2011 [21], and Places365 [14].”
Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning (2020, methodology): “In Appendix F, we assess the generality of BYOL by pretraining a representation on the Places365-Standard dataset [73] before reproducing this evaluation protocol.”
Natural Adversarial Examples (2019, methodology): “For instance, previous work uses the Describable Textures Dataset [9], Places365 scenes [58], and synthetic blobs to test ImageNet OOD detectors.”
Bag of Tricks for Image Classification with Convolutional Neural Networks (2018, background): “In addition, we show that our approach can generalize to other networks (Inception V3 [1] and MobileNet [11]) and datasets (Place365 [33]).”
Resolution-robust Large Mask Inpainting with Fourier Convolutions (2021, methodology): “5M images from Places-Challenge dataset [66].”
Large Scale Image Completion via Co-Modulated Generative Adversarial Networks (2021, methodology): “, 2019a) within a 10k validation split (top two examples) and the Places2 validation set (Zhou et al., 2017) (bottom two examples).”
Regionwise Generative Adversarial Image Inpainting for Large Missing Areas (2019, background): “ular datasets, including faces (CelebA-HQ [16]), and natural scenes (Paris StreetView [17] and Places2 [18]).”
EdgeConnect: Generative Image Inpainting with Adversarial Edge Learning (2019, methodology): “The results over Places2 dataset are reported in Table 1.”
Free-Form Image Inpainting With Gated Convolution (2018, methodology): “We first evaluate our proposed inpainting model on Places2 [Zhou et al. 2017] and CelebA-HQ faces [Karras et al. 2017].”
Learning to Prompt for Vision-Language Models (2021, background): “Unlike existing domain generalization datasets where the domain labels are manually deﬁned and often limited to image style variations, DOSCO-2k focuses on broader contextual domain shift, which is automatically detected by a neural network pre-trained on the Places dataset (Zhou et al., 2017).”
OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (2019, methodology): “For this, we ran a Places2 [52] clas-siﬁer on our images and looked at the top-1 scene class for each image and compared that to COCO overall.”
Towards fairer datasets: filtering and balancing the distribution of the people subtree in the ImageNet hierarchy (2019, methodology): “…the details change: e.g., PASCAL was annotated in-house rather than using crowdsourcing [26], Places relies on both positive and negative verification [84], COCO favors very detailed annotation per image [48], and Open Images [46] and Places [84] both use a computer-assisted annotation approach.”
MaskGIT: Masked Generative Image Transformer (2022, methodology): “We compare MaskGIT against common GAN-based baselines, including DeepFillv2 [52] and HiFill [50], on in-painting with a central 50% ˆ 50% mask, which are evaluated on the Places2 validation set.”
Squeeze-and-Excitation Networks (2017, methodology): “We conduct experiments on the Places365-Challenge dataset [53] for scene classification.”
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization (2016, methodology): “9: Examples showing visual explanations and textual explanations for VGG-16 trained on Places365 dataset [61].”
Learning without Forgetting (2016, methodology): “For the original (“old”) task, we consider the ILSVRC 2012 subset of ImageNet [4] and the Places365-standard [33] dataset.”