Spectral segmentation with multiscale graph decomposition

By Timothée Cour, Florence Bénézit, Jianbo Shi. 2005

643 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

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 Media Technology

Citations in the corpus

Fast Edge Detection Using Structured Forests (2014, neutral): “78 10 Normalized Cuts [10] .”
Pushing the Boundaries of Boundary Detection using Deep Learning (2015, methodology): “Even though accelerations exist, e.g. Cour et al. (2005), we found it simpler to harness the computational power of GPUs and integrate the Damascene system of (Catanzaro et al., 2009) with the Caffe deep learning framework.”
Crisp Boundary Detection Using Pointwise Mutual Information (2014, neutral): “…(Figure 4(a); note that we use decorrelated L*a*b* color space Algorithm ODS OIS AP Canny [14] 0.60 0.63 0.58 Mean Shift [36] 0.64 0.68 0.56 NCuts [37] 0.64 0.68 0.45 Felz-Hutt [38] 0.61 0.64 0.56 gPb [1] 0.71 0.74 0.65 gPb-owt-ucm [1] 0.73 0.76 0.73 SCG [9] 0.74 0.76 0.77 Sketch Tokens [7] 0.73…”
Object detection and segmentation from joint embedding of parts and pixels (2011, methodology): “As in recent effective image segmentation approaches based on spectral clustering [5, 19, 1], we use the intervening con tour cue [10] to define affinities between pixels.”
SLIC Superpixels Compared to State-of-the-Art Superpixel Methods (2012, methodology): “The ability of a superpixel method to adhere to boundaries found in the Berkeley data set [20] is measured and ranked according to two standard metrics: under-segmentation error and boundary recall (for (cid:24) 500 superpixels).”
Matrix Backpropagation for Deep Networks with Structured Layers (2015, methodology): “We use as baselines the original normalized cuts [10] using intervening contour affinities as well as normalized cuts with affinities derived from non-finetuned deep features in different layers of AlexNet (ReLU-5 - the last local ReLU before the fully connected layers) and VGG (first layer in block”
Boundary Learning by Optimization with Topological Constraints (2010, methodology): “We also show that gPb-OWT-UCM [5] and multiscale normalized cuts [6] produce far worse segmentations.”
Segmentation using superpixels: A bipartite graph partitioning approach (2012, methodology): “We compare the average scores of SAS and the ten benchmark algorithms, Ncut [26], Mean Shift [4], FH [10], JSEG [6], Multi-scale Ncut (MNcut) [5], Normalized Tree Partitioning (NTP) [28], Saliency Driven Total Variation (SDTV) [9], Texture and Boundary Encoding-based Segmentation (TBES) [21], Ultram”
Segmentation of Natural Images by Texture and Boundary Compression (2010, neutral): “…whether these problems are unique to our method or are more universal, we have investigated similar problematic cases with the other methods reported here (Arbelaez et al. (2009); Comanicu & Meer (2002); Cour et al. (2005); Felzenszwalb & Huttenlocher (2004); Tu & Zhu (2002); Yang et al. (2008)).”
Recovering Occlusion Boundaries from a Single Image (2007, methodology): “However, most segmentation algorithms rely on 2D perceptual grouping cues, such as brightness, color, or texture similarity for region-based methods [1, 6, 18] or edge strength, continuity, and closure for contour-based methods (e.g., [14]).”
Boundary Detection Benchmarking: Beyond F-Measures (2013, methodology): “For instance, intervene contour [14, 6] is a well-established method that computes the affinity of two points in the image by integrating the boundary strengths along the path that connects those two points.”
Training Deep Networks with Structured Layers by Matrix Backpropagation (2015, methodology): “Arch [47] AlexNet VGG Layer ReLU-5 ReLU-4 ReLU-5 Method NCuts NCuts DeepNCuts NCuts DeepNCuts NCuts DeepNCuts Results .”
Beyond pixels: A comprehensive survey from bottom-up to semantic image segmentation and cosegmentation (2015, background): “Recent work of Cour et al. [48] solves this limitation by deﬁning the afﬁnity matrix at multiple scales and then setting up cross-scale constraints which achieved better result.”
80 Million Tiny Images: A Large Data Set for Nonparametric Object and Scene Recognition (2008, methodology): “To reduce the number of crops that need to be evaluated, we first segment the image using normalized cut s [11], producing around 10 segments (segmentation is perfor med on the high resolution image).”
Object categorization using co-occurrence, location and appearance (2008, methodology): “Stability based image segmentation was implemented with normalized cuts [5, 22], using brightness and texture cues.”