Class segmentation and object localization with superpixel neighborhoods

By Brian Fulkerson, Andrea Vedaldi, Stefano Soatto. 2009

716 citations recorded by the source corpus.

Subfields: 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 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): “Most of the successful semantic segmentation systems developed in the previous decade relied on hand-crafted features combined with flat classifiers, such as Boosting [24], [42], Random Forests [43], or Support Vector Machines [44].”
Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials (2011, background): “Basic CRF models are composed of unary potentials on individual pixels or image patches and pairwise potentials on neighboring pixels or patches [19, 23, 7, 5].”
Feedforward semantic segmentation with zoom-out features (2014, methodology): “The idea of using non-local evidence in segmentation, and specifically of computing features over a neighborhood of superpixels, was introduced in [11] and [25]; other early work on using forms of context for segmentation includes [32].”
Semantic Image Segmentation via Deep Parsing Network (2015, background): “Existing works such as [31, 29, 9, 34, 11, 2, 8, 25, 22] can be generally categorized into two groups based on their definitions of the unary and pairwise terms of MRF.”
Learning a Dictionary of Shape Epitomes with Applications to Image Labeling (2013, methodology): “We demonstrate its quantitative and qualitative properties of our approach with image labeling experiments on two standard datasets: MSRC-21 and Stanford Background.”
Analysis of Hand Segmentation in the Wild (2018, methodology): “6 mIOU), and report possible causes along with failure cases in Fig.5. Performance analysis of CRF. After computing unary and pairwise potentials from test images and classiﬁer results, we used GCMex [12,5,4,19] for energy minimization. We ﬁnd that, although CRF helps to visually improve ﬁnelevel de”
Object Recognition by Sequential Figure-Ground Ranking (2011, background): “2009), or superpixels (Fulkerson et al. 2009; Gonfaus et al. 2010), based on features extracted over a supporting neighborhood.”
CRF Learning with CNN Features for Image Segmentation (2015, methodology): “Different from the binary cases as Weizmann horse and Graz-02, features of the 7th layer perform the best, which may results from the fact that MSRC is much more difficult due to the many categories.”
Stixmantics: A Medium-Level Model for Real-Time Semantic Scene Understanding (2014, background): “(semantic) segmentation approaches, such as [6,20,28,37,44] among many others, operate in a bottom-up fashion and usually do not recover an object-level scene description.”
Advanced Structured Prediction (2014, methodology): “The difference between the Herding approach to integrating locally-trained models and the CRF-based approach adopted in Fulkerson et al. (2010) should now be clear.”
Harmony Potentials (2011, methodology): “Bottom-up image segmentation methods try to label each pixel with the most likely class relying only on local information (Shotton et al. 2009; Yang et al. 2007; Pantofaru et al. 2008; Jiang and Tu 2009; Fulkerson et al. 2009).”
Where's Waldo: Matching people in images of crowds (2011, methodology): “We use the MATLAB implementation of Graph Cuts made available by Fulkerson et al. [ 5 ].”
Video Object Segmentation through Spatially Accurate and Temporally Dense Extraction of Primary Object Regions (2013, methodology): “(2 ∑ (Ci − Cj)(2))|(i,j)∈N We use the graph-cuts based minimization method in [8] to obtain the optimal solution for equation 7, and thus get the final segmentation results.”
Closed-Form Approximate CRF Training for Scalable Image Segmentation (2014, methodology): “In computer vision it is common to use hybrid techniques (e.g. [14, 26, 27]): ﬁrst, ordinary classiﬁers, e.g. SVMs or random forests, are trained to predict unary features.”
Efficient Multi-cue Scene Segmentation (2013, methodology): “While finegrained superpixels have been used as regions [11, 19, 28], recent work shows impressive performance using segmentation trees that contain larger region hypotheses [2, 4].”
The Visual Extent of an Object (2011, background): “These include Local Pixel Context (Carbonetto et al. 2004; Dalal and Triggs 2005; Fulkerson et al. 2009 ; G ould et al.2009; Shotton et al. 2009), 2D scene gist context (Oliva and Torralba 2001), 3D geometric context (Hoiem et al. 2008; Nedovi´ c and Smeulders 2010), and semantic context (Malisiewi”
SLIC Superpixels Compared to State-of-the-Art Superpixel Methods (2012, methodology): “Previous works have used QS08 for object localization [9] and motion segmentation [2].”
Structured Learning and Prediction in Computer Vision (2011, background): “[47] introduce a conditional random field on the superpixels level for the classical problem of figure-ground segmentation, i.”
SLIC Superpixels ? (2010, methodology): “In this section we compare our superpixel segmentation method with four stateof-the-art algorithms, namely, GS041 [8], NC052 [6], TP093 [12], QS094 [7] using publicly available source codes.”