Fast pose estimation with parameter-sensitive hashing

By Gregory Shakhnarovich, Paul A. Viola, Trevor Darrell. 2003

953 citations recorded by the source corpus.

Example-based methods are effective for parameter estimation problems when the underlying system is simple or the dimensionality of the input is low. For complex and high-dimensional problems such as pose estimation, the number of required examples and the computational complexity rapidly become prohibitively high. We introduce a new algorithm that learns a set of hashing functions that efficiently index examples relevant to a particular estimation task. Our algorithm extends locality-sensitive hashing, a recently developed method to find approximate neighbors in time sublinear in the number of examples. This method depends critically on the choice of hash functions that are optimally relevant to a particular estimation problem. Experiments demonstrate that the resulting algorithm, which we call parameter-sensitive hashing, can rapidly and accurately estimate the articulated pose of human figures from a large database of example images.

Subfields: Computer Vision and Pattern Recognition, 3D Vision & Pose

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 3D Vision & Pose

Citations in the corpus

DeepPose: Human Pose Estimation via Deep Neural Networks (2013, methodology): “1, have been recognized and methods reasoning about pose in a holistic manner have been proposed [15, 21] but with limited success in real-world problems.”
Joint Training of a Convolutional Network and a Graphical Model for Human Pose Estimation (2014, background): “Shakhnarovich et al. [31] learn a parameter sensitive hash function to perform example-based pose estimation.”
Human Pose Estimation with Iterative Error Feedback (2015, background): “This is desirable in order to attack problems with higher-dimensional output spaces such as 3D human pose estimation [32, 33] or segmentation.”
Single Image 3D Interpreter Network (2016, methodology): “Training with synthetic data Synthetic data are often used to augment the training set [39,29,36].”
Learning Human Pose Estimation Features with Convolutional Networks (2013, background): “Shakhnarovich et al. [39] learn a parameter sensitive hash function to perform example-based pose estimation.”
Efficient shape matching using shape contexts (2005, methodology): “[28] used a variant of the “Locality Sen sitive Hashing” (LSH) of Indyk and Motwani [29] to quickly retrieve human body shapes.”
Learning invariance through imitation (2011, methodology): “To simplify notation, we use a single superscript, e.g. dmij = d mm ij when points i and j are in the same set as well as drop the superscript on labels within a set: |ymi − ymj | = |yi − yj |.”
Fast Exact Search in Hamming Space With Multi-Index Hashing (2013, neutral): “Nearest neighbor (NN) search on binary codes is used for image search [29], [35], [38], matching local features [2], [7], [16], [33], image classiﬁcation [6], object segmentation [19], and parameter estimation [32].”
Nonparametric Scene Parsing via Label Transfer (2011, neutral): “Ç”
Synthesizing Training Images for Boosting Human 3D Pose Estimation (2016, methodology): “An additional processing stage is then required in order to estimate the 3D pose from 2D joints [45, 6, 17, 40, 12, 4, 57, 55, 52].”
Learning to Hash with Binary Reconstructive Embeddings (2009, neutral): “These include semantic hashing [3], spectral hashin g [4], parameter-sensitive hashing [13], and boosting-based hashing methods [14].”
HumanEva: Synchronized Video and Motion Capture Dataset and Baseline Algorithm for Evaluation of Articulated Human Motion (2010, methodology): “…e.g. average root-mean-squared (RMS) angular error (Agarwal and Triggs 2004a, 2004b, Sminchisescu et al. 2005), normalized error in joint angles (Shakhnarovich et al. 2003), silhouette overlap (Ramanan et al. 2005; Ramanan and Forsyth 2003), joint center distance (Balan et al. 2005; Grauman et…”
FANNG: Fast Approximate Nearest Neighbour Graphs (2016, methodology): “We present a new method for approximate nearest neighbour search on large datasets of high dimensional feature vectors, such as SIFT or GIST descriptors.”
MoDeep: A Deep Learning Framework Using Motion Features for Human Pose Estimation (2014, background): “Shakhnarovich et al. [31] learn a parameter sensitive hash function to perform example-based pose estimation.”
Kernelized locality-sensitive hashing for scalable image search (2009, background): “To make large-scale search practical, vision researchers have recently explored approximate similarity search techniques , where a predictable loss in accuracy is sacrificed in order to allow fast queries even for high-dimensional inputs [25] [24] [12] [16] [4].”
Articulated Pose Estimation Using Discriminative Armlet Classifiers (2013, methodology): “Shakhnarovich et al. [22] recovered the articulated pose of the human upper body by defining parameter-sensitive hash functions to retrieve similar examples from the training set.”
HumanEva: Synchronized Video and Motion Capture Dataset for Evaluation of Articulated Human Motion (2006, methodology): “A variety of statistical [1, 2, 3, 6, 12, 36, 37, 38] as well as deterministic methods [21, 41, 35] have been developed for tracking people from single [1, 2, 7, 12, 15, 20, 21, 25, 26, 27, 29, 37] as well as multiple [3, 6, 9, 36] views.”
80 Million Tiny Images: A Large Data Set for Nonparametric Object and Scene Recognition (2008, background): “estimation of human pose [36]), character recognitio n [3], and object recognition [5], [35].”