Unsupervised modeling of object categories using link analysis techniques

By Gunhee Kim, Christos Faloutsos, Martial Hebert. 2008

126 citations recorded by the source corpus.

We propose an approach for learning visual models of object categories in an unsupervised manner in which we first build a large-scale complex network which captures the interactions of all unit visual features across the entire training set and we infer information, such as which features are in which categories, directly from the graph by using link analysis techniques. The link analysis techniques are based on well-established graph mining techniques used in diverse applications such as WWW, bioinformatics, and social networks. The techniques operate directly on the patterns of connections between features in the graph rather than on statistical properties, e.g., from clustering in feature space. We argue that the resulting techniques are simpler, and we show that they perform similarly or better compared to state of the art techniques on common data sets. We also show results on more challenging data sets than those that have been used in prior work on unsupervised modeling.

Subfields: Computer Vision and Pattern Recognition, Information Systems, Molecular Biology, Statistical and Nonlinear Physics

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 Information Systems

In Molecular Biology

In Statistical and Nonlinear Physics

Citations in the corpus

Unsupervised Visual Representation Learning by Context Prediction (2015, background): “Many other approaches [20, 27, 16] focus on defining similarity metrics which can be used in more standard clustering algorithms; [42], for instance, re-casts the problem as frequent itemset mining.”
Unsupervised Discovery of Mid-Level Discriminative Patches (2012, background): “Since then, there has been a large body of work in the general area of unsupervised object discovery [25,26,27,28,29,30,31].”
Context as Supervisory Signal: Discovering Objects with Predictable Context (2014, background): “Proponents of unsupervised representation learning [1,2,3,4] and unsupervised object discovery [5,6,7,8,9,10,11,12] have long argued that these approaches have the potential to solve two fundamental problems with supervised methods.”
Unsupervised object class discovery via saliency-guided multiple class learning (2012, methodology): “As stated before, several unsupervised approaches have recently been proposed for the object localization and categorization task [26, 15, 22, 18, 20, 38, 19, 32, 21].”
You-Do, I-Learn: Discovering Task Relevant Objects and their Modes of Interaction from Multi-User Egocentric Video (2014, methodology): “Moreover, all these approaches [15, 16, 38] assume the dataset contains a single instance of an object of interest per image.”
Learning the easy things first: Self-paced visual category discovery (2011, methodology): “Existing methods treat unsupervised category discovery as a one-pass “batch” procedure: the input is a set of unlabeled images, and the output is a set of k discovered categories found via clustering or topic models [19, 13, 6, 11, 23].”
Unsupervised Detection of Regions of Interest Using Iterative Link Analysis (2009, background): “Recently, link analysis techniques on visual similarity networks were successfully exploited in computer vision problems [12, 15, 11, 16].”
From a Set of Shapes to Object Discovery (2010, methodology): “For all three graphs, we apply the standard PageRank algorithm, also used in [9,17,10], to identify the most relevant contours, which are then interpreted as object contours.”
Image webs: Computing and exploiting connectivity in image collections (2010, background): “In [15], a lower-level structure is built with graph nodes corresponding to individual local features, with edges linking features sharing a similar descriptor and passing a geometric consistency test.”
Object discovery in 3D scenes via shape analysis (2013, background): “A large portion of these methods focuses on identifying visually similar regions across several images, thereby identifying object classes [4], [5], [6], [7], [8], [9], [10], [11], [12], [13].”
Mining And-Or Graphs for Graph Matching and Object Discovery (2015, methodology): “Object discovery & co-segmentation: Operation 1 in each model-mining iteration can be regarded as a process of object discovery; therefore, we compared the object discovery performance between the mined AoG and six recent methods for object discovery and co-segmentation: saliency-guided multiple cla”
You-Do, I-Learn: Egocentric unsupervised discovery of objects and their modes of interaction towards video-based guidance (2015, background): “Most previous approaches assume the number of groupings is known a priori [25, 57] to avoid the complexity.”
Dataset fingerprints: Exploring image collections through data mining (2015, unknown): “Other graph building approaches are [13, 12].”
Advanced Structured Prediction (2014, methodology): “After linking all pairs of edges and computing their dissimilarities, Page Rank (Brin and Page, 1998; Kim et al., 2008; Lee and Grauman, 2009) iteratively estimates the degree of repetition of each edge as ψi2 = (1− ρ) + ρ ∑”
Unsupervised Object Discovery: A Comparison (2010, methodology): “Finally, Kim et al. (2008) build on link analysis techniques developed in the context of graph-mining and typically used in web search engines.”
Shape discovery from unlabeled image collections (2009, background): “Our patch-anchored shape matching significantly outperforms the baseline using only patch features (note the FPR axes range difference).”
Attributed Graph Mining and Matching: An Attempt to Define and Extract Soft Attributed Patterns (2014, background): “For example, many studies of common object extraction from images [13, 21, 27, 32, 7, 20] use techniques related to maximal clique mining [25, 28] to some extent.”