The variable bandwidth mean shift and data-driven scale selection

By Dorin Comaniciu, Visvanathan Ramesh, Peter Meer. Proceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001, 2001

503 citations recorded by the source corpus.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/a67e0c5b11caceafc0002b0aaff8d7db21eb3312

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

Pedestrian detection in crowded scenes (2005, methodology): “Object hypotheses are found as maxima in the 3D voting space using Mean Shift Mode Estimation [6] with a scale-adaptive balloon density estimator2 [7] and a uniform cubical kernel K:”
The Visual Object Tracking VOT2014 Challenge Results (2014, unknown): “eASMS tracker is a variation of the scale adaptive mean-shift [11,10,12].”
The 2005 PASCAL Visual Object Classes Challenge (2005, methodology): “More precisely, we convert the classifier score to a weight at each 3D point, and use a variable bandwidth Mean Shift algorithm [9] to locate the local maxima of the resulting 3D “probability density”.”
Mid-level Visual Element Discovery as Discriminative Mode Seeking (2013, methodology): “One way to deal with the non-uniformity of the feature space is to use an adaptive bandwidth [4]: that is, different bandwidths are used in different regions of the space.”
Detection and Tracking of Multiple, Partially Occluded Humans by Bayesian Combination of Edgelet based Part Detectors (2007, unknown): “Whenever data association fails (the detectors can not find the object or the affinity is low), a meanshift tracker ( Comaniciu et al., 2001 ) is applied to track the parts individually.”
Optimal Spatial Adaptation for Patch-Based Image Denoising (2006, methodology): “The use of variable bandwidth mean-shift procedures may be also considered for image simplification [ 13 ], but this approach is problematic since it is known to be highly sensitive to noise in images and to the choice of a global initial bandwidth.”
Robust Object Detection with Interleaved Categorization and Segmentation (2008, methodology): “We can thus make the bandwidth dependent on the scale coordinate and obtain the following balloon density estimator (Comaniciu et al. 2001): For K we use a uniform ellipsoidal or cuboidal kernel with a radius corresponding to 5% of the hypothesized object size.”
N-dimensional probability density function transfer and its application to color transfer (2005, background): “A clear outline of the bandwidth selection is available in [2].”
Local adaptivity to variable smoothness for exemplar-based image denoising and representation (2005, unknown): “Also, some efforts have been reported to locally vary the bandwidths in kernel regression [18, 68].”
Mean-shift blob tracking through scale space (2003, methodology): “In [5], methods for adaptive kernel size selection are explored in the context of clustering point sample distributions for image segmentation, however we believe these methods are currently too computationally expensive for real-time tracking applications.”
Automatic detection of microaneurysms in color fundus images (2007, methodology): “One possibility to overcome this problem is to choose the bandwidth depending on z (balloon estimators) or depending on zi (point estimators); the latter ones have been shown to be more efficient (Comaniciu et al., 2001).”
Robust Object Tracking by Hierarchical Association of Detection Responses (2008, background): “Traditional feature-based tracking methods, such as those based on color [1], salient points [2] or motion blobs [3], do not have a discriminative model that distinguishes the object category of interest from others.”
Fast PRISM: Branch and Bound Hough Transform for Object Class Detection (2011, methodology): “In order to cope with scale-invariant object detection, a modified, scaleadaptive version (Comaniciu et al. 2001) of mean-shift had to be used.”
Mean Shift: A Robust Approach Toward Feature Space Analysis (2002, methodology): “We present in [15] a detailed analysis of the bandwidth selection problem.”
Weakly Supervised Learning of Interactions between Humans and Objects (2012, unknown): “Clustering is performed with a weighted dynamicbandwidth mean-shift algorithm based on [27].”
Track and Transfer: Watching Videos to Simulate Strong Human Supervision for Weakly-Supervised Object Detection (2016, background): “Using an adaptive bandwidth [6] to automatically find an optimal value per-image may help to alleviate such errors.”
Automated Mapping of Phenotype Space with Single-Cell Data (2016, background): “In the case of X-shift, we speculate that the elbow point in the cluster number over K plot of X-shift is a result of a balance between bias and variance of the density estimate, which has been thoroughly investigated in the context of kernel density estimation 10 .”
Quick Shift and Kernel Methods for Mode Seeking (2008, background): “If, however, windows kσ(x) of variable width σ are used [6], then the relative weights of the windows become important.”