Learning Deep Features for Scene Recognition using Places Database

By Bolei Zhou, Àgata Lapedriza, Jianxiong Xiao, Antonio Torralba, Aude Oliva. 2014

3,095 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

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 Representation & Self-Supervised Learning

Citations in the corpus

Aligning Books and Movies: Towards Story-Like Visual Explanations by Watching Movies and Reading Books (2015, methodology): “our sentence embedding is not used, VIS: our video-text embedding not used, SCENE: no hybrid-CNN [42] is used in video representation,”
Going deeper with convolutions (2014, methodology): “a Hamming code [13],[38] as in this work, or product quantization code [31],[17].”
ImageNet Large Scale Visual Recognition Challenge (2014, methodology): “ied, there are many errors, making it less suitable for algorithm evaluation. Datasets such as 15 Scenes (Oliva and Torralba, 2001; Fei-Fei and Perona, 2005; Lazebnik et al., 2006) or recent Places (Zhou et al., 2014) provide a single scene category label (as opposed to an object category). The Imag”
Masked Autoencoders Are Scalable Vision Learners (2021, background): “Table 6 studies transfer learning on the iNaturalists [56] and Places [71] tasks (see A.5).”
What makes ImageNet good for transfer learning? (2016, methodology): “Table 3 compares the results on PASCAL-DET when the PASCAL-removed-ImageNet is used for pre-training against the original ImageNet and a baseline of pretraining on the Places [53] dataset.”
Growing a Brain: Fine-Tuning by Increasing Model Capacity (2017, methodology): “…continual, smooth transfer across multiple tasks since we are able to cumulatively increase model capacity as demonstrated in Table 8 AlexNet that is directly trained on the Places dataset with 400 scene categories [59] and ﬁne-tuning a ﬁxed ImageNet VGG16 with higher capacity by a large margin.”
Learning Representations by Maximizing Mutual Information Across Views (2019, methodology): “We evaluate our model using standard datasets: CIFAR10, CIFAR100, STL10 [Coates et al., 2011], ImageNet 1 [Russakovsky et al., 2015], and Places205 [Zhou et al., 2014].”
S4L: Self-Supervised Semi-Supervised Learning (2019, methodology): “This is done by treating the learned model as a fixed feature extractor, and training a linear logistic regression model on top the features it extracts on a different dataset, usually Places205 [42].”
Learning Visual Representations with Caption Annotations (2020, methodology): “We perform these analyses on the Pascal-VOC07 dataset [17] (VOC) for multi-label classification, and ImageNet-1K (IN-1K) [12] and Places205 [69] datasets for large-scale categorization, using the publicly available code of [7] with slight modifications: We apply heavier data augmentations [9] and tr”
Self-Supervised Learning of Visual Features through Embedding Images into Text Topic Spaces (2017, neutral): “ImageNet [21] 79 71 73 75 25 60 80 75 51 45 60 70 80 72 91 42 62 56 82 62 Places [46] 83 60 56 80 23 66 84 54 57 40 74 41 80 68 90 50 45 61 88 63”
Revisiting IM2GPS in the Deep Learning Era (2017, methodology): “Deep learning has been applied successfully to location prediction [36] and other tasks related to our problem: predicting scene type [40], perceptual attributes [6] such as safety, liveliness and geo-informative attributes [15] like GDP, elevation.”
PlaNet - Photo Geolocation with Convolutional Neural Networks (2016, methodology): “On the SUN database [54], Overfeat [47], a CNN trained on ImageNet [12], consistently outperforms other approaches, including global descriptors like GIST and local descriptors like SIFT, and training on the task-specific Places Database [60] yields another significant boost.”
Learning Visual Features from Large Weakly Supervised Data (2015, background): “We note that achieving state-of-the-art results [5, 34, 39, 60] on these datasets requires the development of tailored pipelines, e.g. , using many image transformations and model ensembles, which is out of the scope of this paper.”
Predicting Deep Zero-Shot Convolutional Neural Networks Using Textual Descriptions (2015, background): “The recent success of the deep learning approaches to object recognition is supported by the collection of large datasets with millions of images and thousands of labels [3, 32].”
Webly Supervised Learning of Convolutional Networks (2015, background): “In this paper, we take the ﬁrst step in demonstrating that it is indeed possible to have competitive or even better performance to ImageNet pretrained CNNs by just exploiting web data at much larger scales.”
Emerging Properties in Self-Supervised Vision Transformers (2021, methodology): “We use ImageNet-1k [60] (“Inet”), Places205 [84], PASCAL VOC [24] and Oxford-102 flowers (“FLOWERS”) [46].”
Unified Perceptual Parsing for Scene Understanding (2018, methodology): “We use the validation set of Places-365 [30] containing 36 , 500 images from 365 scenes as our testbed, since the Places dataset contains images from a variety of scenes and is closer to real world.”
Unsupervised Feature Learning via Non-parametric Instance Discrimination (2018, methodology): “Our experimental results show that our method outperforms the state-of-the-art on image classiﬁcation on ImageNet and Places, with a compact 128 -dimensional representation that scales well with more data and deeper networks.”