Unsupervised Learning of Categories from Sets of Partially Matching Image Features

By Kristen Grauman, Trevor Darrell. 2006

197 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Representation & Self-Supervised Learning

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

In Representation & Self-Supervised Learning

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.”
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 Discovery and Tracking in Video Collections (2015, background): “…process a discovery and tracking procedure since ﬁnding the tubes maximizing foreground conﬁdence across videos is akin to unsupervised object discovery [4, 11, 12, 28, 31], whereas ﬁnding the tubes maximizing temporal consistency within a video is similar to object tracking [1, 2, 13, 26, 40, 41].”
Unsupervised object discovery and localization in the wild: Part-based matching with bottom-up region proposals (2015, background): “Many of them [7, 21, 40] are driven by correspondence techniques, e.”
FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery (2018, background): “Early work on unsupervised object discovery [43, 17, 44, 32, 33, 41] use handcrafted features to cluster object categories from unlabeled images.”
From region similarity to category discovery (2011, background): “By contrast, unsupervised approaches require no labeled training data, and merely seek to discover latent structure in the data, e.g., clusters [ 8 , 16, 18, 20] or taxonomies [1, 19].”
Quick Shift and Kernel Methods for Mode Seeking (2008, methodology): “Moreover, compared to [10] we do not try to explicitly separate an object from its background, but we use a simple holistic representation of each image.”
Efficient Classification for Additive Kernel SVMs (2013, background): “There are two important classes of additive kernels used in the computer vision, which we describe next.”
Toward unsupervised, multi-object discovery in large-scale image collections (2020, background): “out ne-tuning the feature extractor. Moreover, our region proposals come with a nice intrinsic structure which can be exploited to improve object discovery performance. Early work on object discovery [12,15,20,27,30] focused on a restricted setting where images are from only a few distinctive object”
Unsupervised object class discovery via saliency-guided multiple class learning (2012, methodology): “They are either restricted to only categorization (no object localization) [22, 18, 38], or have to resort to a separate detector training process using their localization results [15, 19, 21], or only obtain specialized detectors such as chamfer distance based shape templates [20].”
Shape discovery from unlabeled image collections (2009, result): “Our results on benchmark datasets demonstrate the approach can successfully discover shapes from unlabeled images.”
Weakly-supervised Discovery of Visual Pattern Configurations (2014, background): “Existing approaches discover high-level object categories [26, 12, 5], mid-level patches [22, 3, 14], or low-level foreground features [16] by grouping similar visual patterns (i.”
Unsupervised modeling of object categories using link analysis techniques (2008, methodology): “By following the experimental setup proposed by Grauman and Darrell [11], we iterate the same experiment ten times, in which 100 and 75 images per object are randomly picked in each object class for the Caltech-101 and TUD/ETHZ dataset, respectively.”
LabelMe: Online Image Annotation and Applications (2010, background): “A complementa ry line of research tries to avoid the need to annotate images by developing unsupervised learning algorithms [14], [68], [ 4], [83], [16], [62], [51], [71], [18].”
Keywords to visual categories: Multiple-instance learning forweakly supervised object categorization (2008, methodology): “…for each class model, and must resort to a “folding-in” heuristic when used for prediction [11, 25]; the Normalized Cuts 1Our method is unsupervised in the sense that it does not require human input, but we also refer to it as “weakly supervised” since some partitioning is being done by the search…”
Gaussian Processes for Object Categorization (2010, methodology): “A semi-supervised method using normalized cuts to cluster a graph deﬁned by Pyramid Match distances between examples was presented in Grauman and Darrell (2006b), but this method is not probabilistic nor does it provide for an active learning formalism.”
Non-metric affinity propagation for unsupervised image categorization (2007, methodology): “Previous work [20], [21] utilizing SIFT features for image matching and category learning has focussed on the smaller Caltech4 image database or used supervised learning [21].”
50 Years of object recognition: Directions forward (2013, background): “Grauman and Darrell [171] use a “bag-of-features” type of an approach to recognition.”
Unsupervised Visual Representation Learning by Graph-Based Consistent Constraints (2016, background): “Examples include spectral clustering [27], discriminative clustering [25,26], and alternating optimization algorithms [24,23].”