Discriminative Decorrelation for Clustering and Classification

By Bharath Hariharan, Jitendra Malik, Deva Ramanan. 2012

322 citations recorded by the source corpus.

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

Citations in the corpus

A critical analysis of self-supervision, or what we can learn from a single image (2019, methodology): “The negative space was approximated by a multi-dimensional Gaussian by the Exemplar LDA (Hariharan et al., 2012).”
Unsupervised Learning of Visual Representations Using Videos (2015, methodology): “Our performance is significantly better as compared to 24% by ELDA [18] on HOG and 19% by AlexNet with random parameters (our initialization).”
HOGgles: Visualizing Object Detection Features (2013, methodology): “Consider the top detections for the exemplar object detector [9, 15] for a few images shown in Figure 5.”
Joint Unsupervised Learning of Deep Representations and Image Clusters (2016, background): “Deep Representation Learning Many works use raw image intensity or hand-crafted features [9, 18, 19, 23, 42, 50] combined with conventional clustering methods.”
Parsing IKEA Objects: Fine Pose Estimation (2013, methodology): “We will use this linearity later to learn our discriminative classifier.”
Context as Supervisory Signal: Discovering Objects with Predictable Context (2014, background): “However, one final difficulty is that a single cluster initialized by exemplarLDA may actually contain two separate objects, and the verification procedure will happily verify both of them.”
Watch and learn: Semi-supervised learning of object detectors from videos (2015, methodology): “an LSVM on the thousands of boxes, we subsample the labeled boxes (5000 boxes in total for each method using k-means on WHO [23] features).”
Unsupervised Object Discovery and Tracking in Video Collections (2015, background): “Let r t = ( f t , l t ) ∈ R t be a region with its 8 × 8 HOG descriptor f t [7, 14] and its location l t , i.e ., position and scale.”
CliqueCNN: Deep Unsupervised Exemplar Learning (2016, methodology): “The exemplar paradigm has been successfully employed in diverse areas such as segmentation [11], grouping [10], instance retrieval [2, 19], and object recognition [15, 5].”
Unsupervised object discovery and localization in the wild: Part-based matching with bottom-up region proposals (2015, methodology): “We use 8 × 8 HOG descriptors [10, 22] for f to describe the region patches, and center position and scale for l to specify the location.”
Domain-Adaptive Discriminative One-Shot Learning of Gestures (2014, methodology): “Our contributions are: (i) a method for learning gestures accurately and discriminatively from a single positive training example (in the spirit of one-shot learning and exemplar SVMs/LDAs [21, 25]); (ii) a domain adaptation method for human pose and hand shape that enables generalisation to new dom”
Feature representation for statistical-learning-based object detection: A review (2015, methodology): “Categories Typical feature representation methods Engineering -based features Gradient features SIFT [2], PCA-SIFT1 [3], RIFT [4], GLOH1 [5], SURF [6], DAISY [7], GIST1 [8], RIFF [9], WLD [10], HOG [11][12], integral histogram [13], v-HoG [14][15], CoHOG [16], spHOG [17], PQ-HOG [18], WHO1 [19], ECM”
Enriching Visual Knowledge Bases via Object Discovery and Segmentation (2014, methodology): “These overlapping clusters are built using the recent work in training instance based detectors and then using these detectors to find similar instances in the training data [24, 10, 7, 33].”
FPM: Fine Pose Parts-Based Model with 3D CAD Models (2014, methodology): “Note that we do not require annotated images to learn these matrices(2), and they can be efficiently modified for templates with different sizes as described in [17].”
Learning and Calibrating Per-Location Classifiers for Visual Place Recognition (2013, methodology): “While we use hinge loss (Scholkopf and Smola 2002) to train the new discriminative representation of each place, (Aubry et al. 2014; Gharbi et al. 2012; Hariharan et al. 2012) use the Euclidean loss.”
Unsupervised Image Matching and Object Discovery as Optimization (2019, methodology): “where a ij is a similarity term based on appearance alone, using the WHO descriptor (whiten HOG) [13, 19] in our case, r i and r l j denote the image rectangles associated with the two proposals, o is a discretized offset (translation plus two scale factors) taking values in O, and g(r, s, o) measur”
Fisher vectors meet Neural Networks: A hybrid classification architecture (2015, unknown): “[20].”