A latent model of discriminative aspect

By Ali Farhadi, Mostafa Kamali Tabrizi, Ian Endres, D.A. Forsyth. 2009

82 citations recorded by the source corpus.

Recognition using appearance features is confounded by phenomena that cause images of the same object to look different, or images of different objects to look the same. This may occur because the same object looks different from different viewing directions, or because two generally different objects have views from which they look similar. In this paper, we introduce the idea of discriminative aspect, a set of latent variables that encode these phenomena. Changes in view direction are one cause of changes in discriminative aspect, but others include changes in texture or lighting. However, images are not labelled with relevant discriminative aspect parameters. We describe a method to improve discrimination by inferring and then using latent discriminative aspect parameters. We apply our method to two parallel problems: object category recognition and human activity recognition. In each case, appearance features are powerful given appropriate training data, but traditionally fail badly under large changes in view. Our method can recognize an object quite reliably in a view for which it possesses no training example. Our method also reweights features to discount accidental similarities in appearance. We demonstrate that our method produces a significant improvement on the state of the art for both object and activity recognition.

Subfields: Biomedical Engineering, 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 Biomedical Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Histogram of Oriented Principal Components for Cross-View Action Recognition (2014, background): “Keywords: Spatio-temporal keypoint, pointcloud, view invariance.”
HOPC: Histogram of Oriented Principal Components of 3D Pointclouds for Action Recognition (2014, background): “Another approach is to find a view independent latent space [7,8,12,14] in which features extracted from the actions captured at different view points are directly comparable.”
Discriminative virtual views for cross-view action recognition (2012, methodology): “See [5, 25] for more details.”
Cross-view action recognition via view knowledge transfer (2011, background): “Our performance is competitive to that of [8], even though our approach is unsupervised whereas [8] requires supervision.”
A survey on vision-based human action recognition (2010, background): “Recently, researchers have begun to address the recognition of actions from viewpoints for which there is no corresponding training data [33,34,61].”
Learning a Deep Model for Human Action Recognition from Novel Viewpoints (2016, background): “Knowledge transfer-based methods [17], [18], [19], [20], [21], [22], [23], [39] have recently become popular for crossview action recognition.”
Cross-view activity recognition using Hankelets (2012, methodology): “Table 5 gives side by side the average accuracy using BoBHks with the accuracy of previous approaches [16, 7, 10, 6].”
Learning a non-linear knowledge transfer model for cross-view action recognition (2015, background): “Recently, knowledge transfer-based methods [6,7,10,19, 21,40,41] have become popular for cross-view action recognition.”
Cross-View Action Recognition via Transferable Dictionary Learning (2016, background): “THE ACCURACY NUMBERS IN THE BRACKET ARE THE AVERAGE RECOGNITION ACCURACIES OF [12] AND THE PROPOSED SU-RSTDL AND SU-RLTDL RESPECTIVELY”
Learning View-Invariant Sparse Representations for Cross-View Action Recognition (2013, methodology): “The accuracy numbers in the bracket are the average recognition accuracies of [7] and our supervised approach respectively.”
Latent Multitask Learning for View-Invariant Action Recognition (2013, background): “The transformation is learned on the co-occurrence statistics of view-dependent features.”
Depth-Encoded Hough Voting for Joint Object Detection and Shape Recovery (2010, background): “It is superior than [22] and comparable to [7] (which primarily focuses on pose estimation only).”
Learning and Transferring Mid-level Image Representations Using Convolutional Neural Networks (2014, background): “Another enabling factor has been the development of increasingly large and realistic image datasets providing object annotation for training and testing, such as Caltech256 [18], Pascal VOC [11] and ImageNet [9].”
Cross-View Action Recognition via a Continuous Virtual Path (2013, methodology): “Note that we omit the accuracy of [6] and [7], since they report lower results than [17] and [15] in most view combinations.”
Designing Category-Level Attributes for Discriminative Visual Recognition (2013, methodology): “[7,30] use attribute-like latent models to improve object recognition.”