Exploiting hierarchical context on a large database of object categories

By Myung Jin Choi, Joseph J. Lim, Antonio Torralba, Alan S. Willsky. 2010

386 citations recorded by the source corpus.

There has been a growing interest in exploiting contextual information in addition to local features to detect and localize multiple object categories in an image. Context models can efficiently rule out some unlikely combinations or locations of objects and guide detectors to produce a semantically coherent interpretation of a scene. However, the performance benefit from using context models has been limited because most of these methods were tested on datasets with only a few object categories, in which most images contain only one or two object categories. In this paper, we introduce a new dataset with images that contain many instances of different object categories and propose an efficient model that captures the contextual information among more than a hundred of object categories. We show that our context model can be applied to scene understanding tasks that local detectors alone cannot solve.

Subfields: Computer Vision and Pattern Recognition

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

Citations in the corpus

Unsupervised Discovery of Mid-Level Discriminative Patches (2012, background): “For example, “wall” is a well-deﬁned semantic category (with plenty of training data available [21]), but it makes a lousy detector [21] simply because walls are usually plain and thus not easily discriminable.”
Undoing the Damage of Dataset Bias (2012, methodology): “For example, if Caltech101 is the current unseen test set, then the model is trained on PASCAL2007, LabelMe, and SUN09.”
Every Moment Counts: Dense Detailed Labeling of Actions in Complex Videos (2015, methodology): “[2]; we adapt their method to perframe action annotations in video.”
Visual Relationship Detection with Language Priors (2016, background): “Structured learning approaches have improved scene classiﬁcation along with object detection using hierarchial contextual data from co-occurring objects [15,16,17,18].”
Unified Deep Supervised Domain Adaptation and Generalization (2017, methodology): “In this section, we use images of 5 shared object categories (bird, car, chair, dog, and person), of the PASCAL VOC2007 (V) [16], LabelMe (L) [52], Caltech-101 (C) [18], and SUN09 (S) [10] datasets, which is known as VLCS dataset [17].”
Modeling Visual Context is Key to Augmenting Object Detection Datasets (2018, methodology): “To encode more structure, graphical models were then widely used in order to jointly model appearance, geometry, and contextual relations [19,20].”
Reinforcement Learning for Visual Object Detection (2016, methodology): “In contrast to methods based on branch and bound and cascades of classifiers, sequential search methods like [13] attempt to sparsely sample the image through a local search guided by the contextual relations among regions, previously shown to improve detection accuracy[6] [7] [16] [31].”
Contextualizing Object Detection and Classification (2015, methodology): “In this work, the Context-SVM with tractable control on the complexity of the context model is well formulated, so that the generalization capability is guaranteed.”
Domain Generalization Using a Mixture of Multiple Latent Domains (2019, background): “2008), Caltech101 (Fei-Fei, Fergus, and Perona 2007), Sun09 datasets (Choi et al. 2010) which are considered as four separate domains.”
Image Retrieval with Structured Object Queries Using Latent Ranking SVM (2012, methodology): “Instead, we create our own dataset based on the SUN 09 dataset [17] originally collected for the study of scene recognition and object detection.”
Graph Cut Based Inference with Co-occurrence Statistics (2010, background): “As discussed eloquently in the recent work [4], the need for a bias towards parsimony becomes increasingly important as the number of classes to be considered increases.”
Unbiased Metric Learning: On the Utilization of Multiple Datasets and Web Images for Softening Bias (2013, methodology): “One of the forces driving the development is the availability of medium or large size high quality image datasets (e.g., Caltech 101 [7], PASCAL VOC [4], LabelMe [20] and SUN09 [3]) that enable researchers to evaluate and practice sophisticated feature designing and machine learning techniques.”
Exploring Person Context and Local Scene Context for Object Detection (2015, background): “[3] reason both about the spatial relationship of object categories and about the global presence or absence of these categories, thus bringing in global scene context.”
Domain Generalization with Adversarial Feature Learning (2018, methodology): “We then evaluate the performance on other vision problems, such as object recognition on Caltech [8], PASCAL VOC2007 [6], LabelMe [29] and SUN09 [2], as well as the action recognition based on different angles on IXMAS [38].”