Recognizing indoor scenes

By Ariadna Quattoni, Antonio Torralba. 2009

1,785 citations recorded by the source corpus.

Indoor scene recognition is a challenging open problem in high level vision. Most scene recognition models that work well for outdoor scenes perform poorly in the indoor domain. The main difficulty is that while some indoor scenes (e.g. corridors) can be well characterized by global spatial properties, others (e.g, bookstores) are better characterized by the objects they contain. More generally, to address the indoor scenes recognition problem we need a model that can exploit local and global discriminative information. In this paper we propose a prototype based model that can successfully combine both sources of information. To test our approach we created a dataset of 67 indoor scenes categories (the largest available) covering a wide range of domains. The results show that our approach can significantly outperform a state of the art classifier for the task.

Subfields: Computer Vision and Pattern Recognition, Media Technology

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 Media Technology

Citations in the corpus

Growing a Brain: Fine-Tuning by Increasing Model Capacity (2017, methodology): “ns of our networks outperform these baselines, we mainly focus on the width augmented networks (WA-CNN). Tasks and datasets: We evaluate on standard benchmark datasets for scene classiﬁcation: MIT-67 [44], for ﬁne-grained recognition: Caltech-UCSD Birds (CUB) 200- 2011 [48] and Oxford 102 Flowers [2”
Learning Visual Features from Large Weakly Supervised Data (2015, methodology): “…models, we performed transfer-learning experiments on seven test datasets comprising a range of computer-vision tasks: (1) the MIT Indoor dataset [38], (2) the MIT SUN dataset [55], (3) the Stanford 40 Actions dataset [57], (4) the Oxford Flowers dataset [33], (5) the Sports dataset [17], (6)…”
Webly Supervised Learning of Convolutional Networks (2015, methodology): “To further demonstrate the usage of CNN features directly learned from the web, we also conducted scene classification experiments on the MIT Indoor-67 dataset [37].”
CNN Features Off-the-Shelf: An Astounding Baseline for Recognition (2014, methodology): “We use two challenging recognition datasets, Namely, Pascal VOC 2007 for object image classification [12] and the MIT-67 indoor scenes [36] for scene recognition.”
YouTube-8M: A Large-Scale Video Classification Benchmark (2016, background): “[38, 28] It is also different than most object recognition tasks, where the goal is to label everything visible in an image.”
Learning Deep Features for Discriminative Localization (2015, background): “3.2 (i.e., thresholding) to first identify bird bounding boxes in both the train and test sets.”
Unsupervised Learning of Visual Representations Using Videos (2015, methodology): “We also evaluate our unsupervised-CNN without finetuning for scene classification task on MIT Indoor 67 [36].”
Variational Information Distillation for Knowledge Transfer (2019, background): “The second task is an indoor scene classification based on the MIT-67 dataset [20], which contains 15,620 images for 67 classes of indoor scenes.”
Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer (2016, methodology): “A method combining both guided backpropagation and CAM is Grad-CAM by Selvaraju et al. (2016), adding image-level details to class-discriminative attention maps.”
NAS evaluation is frustratingly hard (2019, methodology): “…CIFAR10 we select CIFAR100 for a more challenging object classiﬁcation problem (Krizhevsky, 2009); SPORT8 for action classiﬁcation (Li & Fei-Fei, 2007); MIT67 for scene classiﬁcation (Quattoni & Torralba, 2009); and FLOWERS102 for ﬁne-grained object classiﬁcation (Nilsback & Zisserman, 2008).”
Interpreting Deep Visual Representations via Network Dissection (2017, positive): “We include indoor67 [53], sun397 [54] and caltech256 [55].”
DLCFT: Deep Linear Continual Fine-Tuning for General Incremental Learning (2022, methodology): “Additionally, we used Seq-MIT-67 for large-resolution dataset, which is MIT-67 dataset [35] split into 4 tasks.”
Memory Aware Synapses: Learning what (not) to forget (2017, methodology): “We consider a sequence of two tasks based on three datasets: MIT Scenes [18] for indoor scene classification (5,360 samples), Caltech-UCSD Birds [30] for fine-grained bird classification (5,994 samples), and Oxford Flowers [16] for finegrained flower classification (2,040 samples).”
Encoder Based Lifelong Learning (2017, methodology): “0.1 Tested scenario We test our method and LwF on a two-task scenario starting from ImageNet (LSVRC 2012 subset) [22] (more than 1 million training images) then training on MIT Scenes [19] for indoor scene classiﬁcation (5,360 samples).”
Expert Gate: Lifelong Learning with a Network of Experts (2016, methodology): “We start with the sequential learning of three image clas-siﬁcation tasks: in order we train on MIT Scenes [26] for scene classiﬁcation, Caltech-UCSD Birds [35] for ﬁne-grained bird classiﬁcation and Oxford Flowers [25] for ﬁne-grained ﬂower classiﬁcation.”
Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (2020, background): “Next, we use several ﬁne-grained classiﬁcation datasets: Oxford Flowers [102], MIT Indoor Scenes [103], CUB-200-2011 Birds [104], Stanford Cars [105], FGVC Air-craft [106], and Stanford Actions [107].”
ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression (2017, background): “These strategies are compared on two different domainspecific dataset: CUB-200 [31] for fine-grained classification and Indoor-67 [25] for scene recognition.”