Fast High‐Dimensional Filtering Using the Permutohedral Lattice

By Andrew Adams, Jongmin Baek, M. A. Davis. 2010

495 citations recorded by the source corpus.

Subfields: Computational Theory and Mathematics, Computer Vision and Pattern Recognition

How researchers see this paper

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

In Computational Theory and Mathematics

In Computer Vision and Pattern Recognition

Citations in the corpus

DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (2016, background): “High-dimensional filtering algorithms [84] significantly speed-up this computation resulting in an algorithm that is very fast in practice, requiring less that 0.”
Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs (2014, methodology): “High-dimensional filtering algorithms (Adams et al., 2010) significantly speed-up this computation resulting in an algorithm that is very fast in practice, less that 0.”
Conditional Random Fields as Recurrent Neural Networks (2015, methodology): “Following [29], we use the Permutohedral lattice implementation [1], which can compute the filter response in O(N) time, where N is the number of pixels of the image [1].”
Rethinking Atrous Convolution for Semantic Image Segmentation (2017, methodology): “One effective method is to incorporate DenseCRF [40] (with efﬁcient high-dimensional ﬁltering algorithms [2]) to DCNNs [9, 10].”
Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials (2011, methodology): “structure that tiles the feature space with simplices arranged along d+1 axes [1].”
Semantic Image Segmentation via Deep Parsing Network (2015, background): “(7) is accelerated by fast Gaussian filtering [1].”
Large Kernel Matters — Improve Semantic Segmentation by Global Convolutional Network (2017, methodology): “Then CRFAsRNN [37] models the denseCRF into a RNN-style operator and proposes an end-to-end pipeline, yet it involves too much CPU computation on Permutohe-dral Lattice [1].”
3D fully convolutional network for vehicle detection in point cloud (2016, background): “2) Data Encoding: Although a variety of discretization embedding have been introduced for high-dimensional convolution [1], [12], for simplicity we discretize the point cloud on square grids.”
MaskLab: Instance Segmentation by Refining Object Detection with Semantic and Direction Features (2017, neutral): “[77] propose a novel fully connected CRF [39] (with fast inference by permutohedral lattice [2]) to refine the results.”
Instance-Level Segmentation for Autonomous Driving with Deep Densely Connected MRFs (2015, methodology): “Following [15], we use the efficient permutohedral lattice data structure [1] to perform downsampling,”
Efficient multi‐scale 3D CNN with fully connected CRF for accurate brain lesion segmentation (2016, background): “This way, the features learned by the ﬁrst pathway retain ﬁnest details, as they are not involved in processing low resolution context.”
On Regularized Losses for Weakly-supervised CNN Segmentation (2018, methodology): “We implement our loss layers using fast Gaussian filtering [16], which is also utilized in the inference of DenseCRF [8,9].”
Zoom Better to See Clearer: Human Part Segmentation with Auto Zoom Net (2015, background): “The reason is that since the fully connected CRF requires bilateral kernels (Adams et al., 2010) for fast inference, it is difficult to recover the human part boundaries with features, such as position and color intensity.”
Salient object detection: A survey (2014, methodology): “[27] demonstrate that if Dr(ri, rj) is defined as the Euclidean color distance between ri and rj , global contrast can be computed using efficient filtering based techniques [88].”
Pixel-Adaptive Convolutional Neural Networks (2019, methodology): “First, since data projected on a higher-dimensional space is sparse, special lattice structures and hash tables are needed to perform the convolution [1] resulting in considerable computational overhead.”
The Fast Bilateral Solver (2015, methodology): “There are techniques for speeding up bilateral filtering [1, 6] which treat the filter as a “splat/blur/slice” procedure: pixel values are “splatted” onto a small set of vertices in a grid [2, 6] or lattice [1] (a soft histogramming operation), then those vertex values are blurred, and then the filt”
A Fully-Connected Layered Model of Foreground and Background Flow (2013, background): “They show that the spatial message passing step can be efficiently approximated by high-dimensional filtering [1].”