Unsupervised Object Discovery: A Comparison

By Tinne Tuytelaars, Christoph H. Lampert, Matthew B. Blaschko, Wray L. Buntine. 2010

221 citations recorded by the source corpus.

The goal of this paper is to evaluate and compare models and methods for learning to recognize basic entities in images in an unsupervised setting. In other words, we want to discover the objects present in the images by analyzing unlabeled data and searching for re-occurring patterns. We experiment with various baseline methods, methods based on latent variable models, as well as spectral clustering methods. The results are presented and compared both on subsets of Caltech256 and MSRC2, data sets that are larger and more challenging and that include more object classes than what has previously been reported in the literature. A rigorous framework for evaluating unsupervised object discovery methods is proposed.

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

Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (2019, background): “Recent work on Neural Expectation Maximization (Greff, van Steenkiste, and Schmidhuber 2017; van Steenkiste et al. 2018) aims to achieve something similar to AIR, but has comparable scaling issues.”
Maximum Margin Temporal Clustering (2012, background): “This is referred as purity in information theoretic measures (Meila, 2007; Tuytelaars et al., 2009).”
Recognizing scene viewpoint using panoramic place representation (2012, methodology): “Similar to Unsupervised Object Discovery [15], we use an oracle to assign each aligned viewpoint to one of the m view directions from the truth, and then evaluate the accuracy based on the resultant viewpoint-to-viewpoint mapping.”
Unsupervised Joint Mining of Deep Features and Image Labels for Large-Scale Radiology Image Categorization and Scene Recognition (2017, methodology): “Two convergence measurements have been adopted from [56], i.”
Category Modeling from Just a Single Labeling: Use Depth Information to Guide the Learning of 2D Models (2013, background): “Unsupervised object discovery (reviewed by [24]) was a classical achievement of object-level knowledge mining.”
Learning Sparse FRAME Models for Natural Image Patterns (2014, methodology): “Experiment 5: Numerical evaluation on clustering To evaluate the clustering accuracies, we use two measures: conditional purity and conditional entropy (Tuytelaars et al. 2009).”
Inducing wavelets into random fields via generative boosting (2016, unknown): “To quantize the clustering accuracies, we use two metrics: conditional purity and conditional entropy [21].”
Weakly supervised semantic segmentation with a multi-image model (2011, background): “Towards less supervision, the extreme case is unsupervised object discovery, where unlabeled images of multiple classes are given [23].”
Unsupervised object class discovery via saliency-guided multiple class learning (2012, background): “Recently, many unsupervised approaches have also been proposed for the object localization and categorization tasks [26, 15, 22, 18, 20, 38, 19, 32, 21].”
Weakly supervised histopathology cancer image segmentation and classification (2014, methodology): “Unsupervised learning methods (Duda et al., 2001; Loeff et al., 2005; Tuytelaars et al., 2009), on the other hand, ease the burden”
Functional categorization of objects using real-time markerless motion capture (2011, methodology): “While unsupervised categorization of objects using an appearance-based similarity measure has been of particular interest in the last years [31], categorizing based on functional similarity in an unsupervised fashion, which is addressed in this work, has received little attention.”
Object Discovery: Soft Attributed Graph Mining (2016, background): “Some pioneering studies, including those on object discovery [30] and co-segmentation [31], [32], [33], can be thought of as a kind of object-level visual mining.”
Object discovery in 3D scenes via shape analysis (2013, unknown): “state-of-the-art techniques can be found in [16].”
You-Do, I-Learn: Discovering Task Relevant Objects and their Modes of Interaction from Multi-User Egocentric Video (2014, background): “Moreover, all these approaches [15, 16, 38] assume the dataset contains a single instance of an object of interest per image.”
Enriching Visual Knowledge Bases via Object Discovery and Segmentation (2014, background): “In order to learn priors with little or no annotations, recent approaches have also tried using object discovery to extract segments from images automatically, followed by learning of priors (see [35] for an overview).”